William Sayers

2009
Milestone 3

05025397

Genetic Algorithms and Neural Networks

Genetic Algorithms and Neural Networks
William Sayers (05025397)
Supervised by: Colin W. Morris (BSc, MSc)
Backpropagation neural networks are usually trained using some iterative method derived from a mathematical analysis of the operation of the network. This technique is time consuming and requires understanding of the mathematical principles involved. This project will investigate the application of genetic algorithms to replace the "normal" training of the network. This involves setting up some candidate networks and then picking the best of these. The best of the population are then combined in some way to produce a new population of candidate networks. This procedure is continued until a satisfactory network is obtained.

William Keith Paul Sayers225 Page 1 of Faculty of Advanced Technology 22/04/2009

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3

Contents
Contents ............................................................................................................................ 2 1 - Introduction ............................................................................................................... 10 2 - Research .................................................................................................................... 10 2.1 - Neural Networks ................................................................................................. 10 2.1.1 - The Biological Neuron................................................................................. 10 Figure 1 - A human neuron ............................................................................................ 10 2.1.2 - The Artificial Neuron................................................................................... 11 Figure 2 - An artificial neuron ........................................................................................ 12 Equation 1 - Summation of the inputs and weights ........................................................ 12 Figure 3 - The sigmoid function shape ........................................................................... 13 2.2 - The History of Neural Networks ........................................................................ 13 2.3 - Advantages and Disadvantages of Neural Networks ......................................... 14 2.3.1 - Advantages................................................................................................... 14 2.3.2 - Disadvantages .............................................................................................. 15 2.4 - Current Applications of Neural Networks .......................................................... 17 2.4.1 - Neural Networks in Medicine ...................................................................... 17 2.4.2 - Neural Networks in Business ....................................................................... 18 2.4.3 - Object Trajectories....................................................................................... 19 2.4.4 - Robot Control .............................................................................................. 19 2.5 - Back-propagation ............................................................................................... 20 2.5.1 - Back-propagation overview ......................................................................... 20 Page 2 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3 2.5.2 - Back-propagation in depth ........................................................................... 20 2.5.3 - Back-propagation library ............................................................................. 22 2.6 - The FANN neural network library (Nissen, FANN) .......................................... 22 2.7 - Genetic Algorithms ............................................................................................ 22 2.7.1 - History of Genetic Algorithms .................................................................... 23 2.7.2 - Advantages and Disadvantages of Genetic Algorithms .............................. 24 Figure 4 - local optimum (units are undefined in this case). .......................................... 27 2.8 - Training Neural Networks with Genetic Algorithms ......................................... 28 2.8.1 - Determining Weight Values with Genetic Algorithms................................ 28 2.8.2 - Representation of a Neural Network within a Genetic Algrithm ................ 28 2.8.3 - Using Genetic Algorithms to Determine Neural Network Structure ........... 29 Figure 5 - A feed-forward neural network ..................................................................... 30 2.9 - Cascade Correlation............................................................................................ 32 2.10 - C# User Interface Programming ....................................................................... 37 2.11 - Writing C++/CLI Wrappers for Unmanaged C++ code................................... 37 2.12 - Application of Research ................................................................................... 38 3 - Design ....................................................................................................................... 38 3.1 - Program Requirements ....................................................................................... 39 3.2 - Design of the Class Structure ............................................................................. 39 3.3 - Linking C# code to managed dll’s...................................................................... 40 3.4 - Design of the User Interface ............................................................................... 40 Page 3 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3 3.4.1 - Main Form ................................................................................................... 40 Figure 9 – Network and output tab design (Main form) ................................................ 41 Figure 10 - Mean squared error tab design (Main form) ................................................ 42 Figure 11 - Dataset tab design (Main form) ................................................................... 42 3.4.2 - New Network Wizard .................................................................................. 43 Figure 12 - Design of the training algorithm tab (Wizard form) .................................... 43 Figure 13 - Design of the network settings tab (Wizard form) ...................................... 44 Figure 14 - Dataset selection tab design (Wizard form)................................................. 44 3.4.3 - Working Form ............................................................................................. 44 Figure 15 - Working form design (Working form) ........................................................ 45 3.4.4 - About Form .................................................................................................. 45 Figure 16 - About form design (About form) ................................................................. 45 Figure 17 - Dataset display design (About form) ........................................................... 46 Figure 18 - Licenses display design (About form) ......................................................... 46 4 - Implementation ......................................................................................................... 46 4.1 - User Interface Implementation ........................................................................... 47 4.1.1 - Main Form ................................................................................................... 47 Figure 19 - Network and Output tab............................................................................... 47 Figure 20 - Mean squared error graph tab ..................................................................... 48 Figure 21 - The dataset tab ............................................................................................. 49 4.1.2 - New Neural Network Wizard ...................................................................... 50 Figure 22 - Learning Algorithm tab ............................................................................... 50 Figure 23 - Network Settings tab.................................................................................... 51

Page 4 of 225

..........Working Form ....................................................3 ....................1 ..................................2..................................William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Figure 24 ....................... 55 4........1...............2 .......................... 60 5....................4 ..........................XOR .......................................................................The mean squared error at testing (XOR)....................................... 59 5........................... 54 4...............................................Back-propagation .............. 54 4...........................................Testing Data .1 ....................Graphs .............................................1.................................................1 ...... Licenses view ... 54 Figure 28 ........................5 ..................1...............................Neural Network Implementation ............About form..... 56 4.......................... 61 Figure 31 .3 ....................Back-propagation and Cascade Correlation Implementation ................................. 57 5 .... Datasets view ......................Number of iterations to achieve target (XOR) ......3 ..........About Form .......4 ......1............................................................................................................... 54 4.............Passing Information between Forms ................Genetic Algorithms ... 52 Figure 26 .............................................................................. 59 5.Dataset tab . 53 Figure 27 .....................................Working form ...........................................1................................. 57 4........................4 ............. 58 5..... 51 4..................... main view ..1.........................Genetic Algorithm Implementation ....................... 58 5............................2 ..........................................Data Parser Implementation .................................1.................... 61 Figure 29 ........... 52 Figure 25 .................................................................................Keeping the User Interface Active whilst Processing is Occurring.6 .......Cascade Correlation .............About form....................1........................... 61 Figure 30 ....................About form... 56 4...........Mean squared error upon last training test (XOR) ...................................... 62 Page 5 of 225 ....................................

..1 .... 69 6..................................................... 68 6...................The mean squared error at testing (Iris data) ...Number of iterations to achieve target (Viruses) ............................. 68 6..................................................1.........Fishers Iris data......................................................................2............Fishers Iris Data .........Virus Classification ...................... 64 Figure 34...................... 66 5..............Graphs .................2........1 ...........2............3 .................................................... 63 5...........3..Back-propagation ............2 ................................................................................XOR ......................................................................1 .....................................................Cascade Correlation and Genetic Algorithms .................Comparisons ......................................................Cascade Correlation .........1...........3 ..4 .................................2................................................. 69 6............ 63 5.......................................... 63 5....... 66 5.............2 ....................... 67 Figure 36 ......... 66 5..........................Back-propagation and Genetic Algorithms ...................................Number of iterations to achieve target (Iris data) .2 ............................................ 68 6 .................Mean squared error at testing (Viruses).......................3.......3 .............................................Back-propagation .3 ..........................................................2 ....Mean squared error upon last training test (Viruses). 65 5.....................................1 .............................................Graphs ......................................4 .............3.................................................................Mean squared error upon last training test (Iris data) ................. 67 Figure 37 ... 64 Figure 32 .... 63 5.. 64 Figure 33.........................Cascade Correlation ........................ 69 Page 6 of 225 ................Genetic Algorithms ....................................... 65 5....................Virus Classification..3. 68 6......................................................................1.........................William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 5........ 67 Figure 35 .....................................................Genetic Algorithms .................................2 .

..........................................Controller ..Works Cited ....................2..Libraries Researched ......................................................1.......Objectives ...........................1 .................... 69 6..............Model ................................................... 71 8 ...How many of the original objectives were achieved? ................................................................................................................................................................ 75 9..........................4 .............................................2 ....Virus Classification.........................................................................................................2 ................... 70 6............ 74 9...........................................................................5 ...... 70 7 ...Fishers Iris Data ...................................DLL Wrapper for FANN functions ................................. 207 9.................... 73 9 .....................................................................1 ............................................2 .... 174 9........... 73 8.......UML Class Diagram..........................1 ........ 74 9..........Possible Training Algorithm Improvements ....................................................................................... 70 7............................................. 209 9.....................................Possible Improvements ..................Possible User Interface Improvements . 211 11 ........ 158 9...........................View ...... 72 8................................................................ 69 6......................2.................................4 .... 210 10 ......Evaluation .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 6.................................Testing conclusions ...................................1........................2............................................1................................................................Appendices ...............1 ....Bibliography...............................................3 ...................Datasets............... 214 Page 7 of 225 ...................................................................................1.............................................................XOR .................3 .......Source code ........................... 109 9...............1 .............................3 ....................................3 .............................................2 ............ 204 9......................

....... 219 12...................................................................... 222 Page 8 of 225 ..........William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 12 .....3 ................................................................Table of Equations ......Tables .......................Table of Figures ..................................................................................................................................................................1 .... 220 13 ..................................................................Diary........2 ... 217 12...................... 217 12...Table of tables ......................................................................

...... Page 9 of 225 ........ nor any part of it..... Any material taken from published texts or computerised sources have been fully referenced.... Student Name (Printed) Student Signature Registered Scheme of Study Date of Signing ………………………………............................ ………………………………........... the work described within this project is the result of the investigation carried out by myself.......... has been submitted in candidature for any other award other than this being presently studied............. except where specific reference is made.........................William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 STATEMENT OF ORIGINALITY SCHOOL OF COMPUTING DEGREE SCHEME IN COMPUTING LEVEL THREE PROJECT This is to certify that.. and I fully realise the consequences of plagiarising any of these sources.............. ………………………………..... ………………………………............ and that neither this project..

In addition. Figure 1 . compare and contrast three common learning algorithms that are applicable to neural networks.1 .A human neuron Page 10 of 225 . Although they are simple in concept.1.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 1 . This report encompasses design.1 . 2 . 2.Neural Networks Neural networks are networks of small-specialised units that sum the products of their inputs and the weights on those inputs. but in combination with a network of other neurons within the brain.The Biological Neuron The human brain is composed of many millions of nerve cells called “Neurons”.Introduction The purpose of this report is to bring to a conclusion my final year project on neural networks and genetic algorithms. on the course “Computer Games Development”.Research 2. the report will investigate. Neurons do not make decisions alone. they can produce immensely complex results. development and final prototyping of the program developed for this project.

The axon and dendrites both split before terminating into “Axon branches” and “Synapses”. (Barber. the nucleus will generate an “action potential”. Concerning inputs from the dendrites.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 A neuron is composed of a cell body. in this gap between neurons chemical reactions take place that will inhibit or excite the signal. Axon branches can connect to other types of cells also. 1. a number of weights associated with those inputs. This action potential extends down the axon to the cells at the end of the axon branches (often synapses connecting to other neuronal cells. or to another part of the program. 2004). as mentioned above). Page 11 of 225 . the nucleus of the neuron cell processes them and produces an output. AI: Neural Network for Beginners pt. a summing function.2 . If the inputs from the dendrites to the neuron nucleus when summed are past a certain threshold.The Artificial Neuron In order to construct an artificial neuron we need a way of representing the same model in data structures and algorithms. 2006) 2. In the case of the axon branches connecting to synapses that are part of a second neuron. in order to allow the brain to control cells external to the brain. a synaptic gap occurs between the axon branch and the synapse. with extensions called “Dendrites” which are the neuronal inputs and a single extension called an “Axon” which is the output for that neuron.1. (Bourg & Seeman. an activation function and a method for representing an output – either to another artificial neuron in the next layer. The usual method for approaching this is to have a number of inputs. Axon branches connect to synapses to form the neural networks within the brain.

Alternatively.Summation of the inputs and weights The next stage depends on the activation function chosen. The product of the data item and its associated weighting is then stored before the summation of all the data items and associated weights occurs. or one (activated) if the summed products are above a certain value. Page 12 of 225 . or integer). A stepping function can be used which will output zero (or non-activated) if the summed products are below a certain value. real number. The activation function accepts this stored value. a differentiable function of some kind (such as a sigmoid function) which will output values between zero and one depending on the value of the inputs is another possibility. A specific input accepts a data item (usually a binary digit.An artificial neuron The system devised with these components is as follows.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Figure 2 . ℎ = × ℎ Equation 1 .

The History of Neural Networks The foundation of neural network research are in psychology. McCulloch and Pitts formulated the first neural network model which featured digital neurons but the neurons had no capability to learn (McCulloch & Pitts.8 0. Frank Rosenblatt developed the perceptron model then. however he was unable to come up with a reliable mathematically accurate mechanism for allowing multilayer perceptrons to learn (Rosenblatt. in 1958.6 0.4 0. Page 13 of 225 . 1943). AI: Neural Network for Beginners pt.2 0 -6 -4 -2 0 2 4 6 Figure 3 . 1. contributed to the ideas that underpinned early neural network research.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 1. as defined in the graph above. (Barber. 1958). William James and others. In this example “y” is the function output and “t” is the function input. 2004). 2006) 2.2 . (Bourg & Seeman. 2004).The sigmoid function shape = 1 1+ Equation 2 .2 1 0.A sigmoid function (Bourg & Seeman. the work of 19th century psychologists such as Freud.

(Blum. 2.3 . 1982)). Back-propagation allows for the training of multilayer perceptrons and in particular. The next major advancement in neural networks was not until 1974. perceptrons with “n” hidden layers. Werbos discovered the algorithm whilst working on his doctoral thesis in statistics and called it “dynamic feedback”.Advantages Neural networks make possible or practical many things that with conventional solutions would be extremely difficult. 1992). 1974) discovered the back-propagation algorithm (independently rediscovered by Parker in 1982 (Parker. of “n” nodes. Parker discovered it independently in 1982 whilst performing graduate work at Stanford University in the United States of America and called it “learning logic”. when Werbos (Werbos. where “n” is an undefined number. A good example of this is that neural networks have been implemented to run on architecture originally designed for processing three dimensional computer graphics. taking Page 14 of 225 . used on this architecture they become an extremely fast as well as flexible tool. One of the benefits of neural networks when compared with other problem solving techniques in that they are inherently parallel and thus can run very effectively and efficiently on parallel hardware.Advantages and Disadvantages of Neural Networks 2.1 .3.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Two layer perceptrons were the subject of several experiments to determine their usage and effectiveness. The individual neurons in a neural network have no time dependencies on each other and each can therefore be run on a separate processor (or as separate threads on a single processor) if desired without causing the problems often associated with such parallelism.

2 Formulate a statistical model. Table 1 . as irrelevant data will simply end up weighted so that it has zero or close to zero actual effect on the solution produced by the network.3. 1 Decide on relevant data. noisy data is utilised during training upon occasion.The statistical approach to deciding relevant data”. 3 Run the formulated model. This is a large disadvantage in areas such as medical diagnosis. in fact. to attempt to provide a more robust post-training network. 2. When using a neural network there's no need to establish before attempting problem solving which data is relevant. Page 15 of 225 . 5 Build a system that incorporates what we have learned. We can simply present the neural network with all the data as opposed to following the circuitous and overly complex (by comparison) statistical approach demonstrated in “Table 1 . 2001). where explaining why you have arrived at a particular diagnosis is an important part of the process.The statistical approach to deciding relevant data Another advantage of neural networks is that noisy data is not a real problem for them to learn to interpret. 4 Analyze the results.2 .Disadvantages One disadvantage to neural networks can be that it is very difficult for a human being to analyse a trained neural network and describe why it may come up with one answer over another. Neural networks can excel at determining what data in a particular set is relevant.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 advantage of the massively parallel architecture of this hardware and the programmable sections of the graphics pipeline (vector and fragment shaders) to produce extremely fast neural networks (Davis.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Once a neural network’s training is complete it is hard to tell why it is coming up with a particular solution. All the photographs with tanks were from a cloudy day and all the images without. they tested the network with the remaining fifty images that the neural network had not seen before and the neural network correctly identified most of the tanks. The US military then tested the neural network with a further set of one hundred images. They collected one hundred images with a concealed tank. a sunny day. Eventually the original data set was re-examined. the US military wanted to use a neural network to analyse images and determine whether they contained a concealed tank. asked to differentiate between the two sets of pictures. A tool used to solve heuristic problems. the network came up with apparently random answers. After the neural network’s training was complete. then a further one hundred images without. train the neural network. only to find that on this round of testing. Fifty pictures of each set were set aside and the neural network trained with the remaining one hundred images with and without tanks. although the length of time the problem would take to solve without employing a neural network must be taken into account when deciding how much of a disadvantage. The training times on neural networks can also be a disadvantage. Therefore. with and without tanks. there is no formal methodology with which to choose the architecture. Page 16 of 225 . requires largely heuristic construction and preparation. for obvious purposes. it can also be hard to tell how well (or how badly) trained a neural network is. A classic example is that in the 1980s. or even to verify the quality of the training. the network had trained itself to recognise whether the sky was cloudy or not. When using neural networks is that.

Current Applications of Neural Networks 2. Neural networks make it more apparent however because (as in the previous US Military example) it can be difficult sometimes to decide what constitutes good data. This is generalisation and it is far more useful than it at first appears and similar from a very high-level perspective to how the human brain and brains in general function. lies in their ability for generalisation. One of the main research areas is using neural networks to recognise diseases/conditions via scans. 2. the eventual result will be. or it does exacerbate the problem with the time spent training the neural network. Testing may then take place with the “testing” set to determine accuracy with unseen data. The more time spent training the neural network on quality data sets. The real strength of neural networks. the more accurate and in line with expectations and desires. Garbage out”. This drawback is however one that is generally universal to computing (and many other areas) often referred to as “GIGO” or “Garbage in.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Another possible criticism is that neural networks are extremely dependant on the quality and content of the original training data set. this has to be done within reason however.4 . Data sets for training neural networks generally separate into two sub-sets. The network training first takes place using the appropriate “training” set.1 . the usual best approach is to simply feed them as much data as can be obtained.Neural Networks in Medicine There is a large amount of research now covering the applications of neural networks in medicine. “training” and “testing”. Fortunately since neural networks are (as has been mentioned above) good at ignoring irrelevant data. in the opinion of the author.4. Page 17 of 225 . covering as many relevant situations as possible.

In the 1980s. (Stergiou & Dimitrios. 1997) Page 18 of 225 . Businesses have used neural networks in the past for applications such as assisting the marketing control of airline seat allocations in the AMT (Airline Marketing Tactician). for example the human cardio-vascular system. (Stergiou & Dimitrios. mortgage screening. A back-propagation neural net is integrated with the airline marketing tactician which monitored and recommended on booking for each flight. thus supplying information more or less directly linked to the airlines main income. However. a clear problem with this approach is that when using a neural network it would be difficult to tell how a neural network reached its conclusion. merely a variety of scans covering all possible permutations of the disease or condition. The main reasons for using a neural network to model the human cardio-vascular system are that a neural network is capable of adjusting itself to be relevant to a particular individual and also to adapt by itself to changes in that individual. 1997) 2. can fit very well into most business situations.2 . being good at analysing patterns and predicting future trends. research on using neural networks to diagnose diseases took place.Neural Networks in Business Neural networks.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 This is particularly suitable work for neural networks because you do not need perfect examples of how to recognise the diseases and conditions. You can also use the technology to model parts of the human biological system in order to better understand it.4. Credit scoring. and assessing borrowers have all incorporated neural networks as an integral part of the system.

When controlling manipulators a neural net must deal with four major problems (Robot Control. Adjust the arm to the appropriate joint angles and close the effecter. (relatively speaking) involving the following steps (Robot Control. velocity and acceleration of the object being estimated in those implementations by several neural networks using several of the most recent measurements of the object coordinates. 2008): • • • • Forward Kinematics Inverse Kinematics Dynamics Trajectory generation The ultimate goal of course is to position the “effecter” (I.4 . A neural network allows the robotic arms to be very flexible in their operation and perform self-adjustments as time goes by. 2008): • • • Determine the target coordinates relative to the robot.3 .4. Calculate the joint angles required for the arm to be in the appropriate position. The position.4. (Marshall & Srikanth. 1999) 2.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 2.Robot Control Controlling manipulators (such as robotic arms in a car manufacturing plant) is usually how robotics and neural networks connect. When a robot arm is one hundred percent accurate that is a simple task.e. a robotic hand or any other similar manipulator) in the appropriate position to grasp or otherwise manipulate an object or part of an object. compensating automatically for wear and tear on Page 19 of 225 .Object Trajectories Predicting the trajectory of an object in real time is another area where neural networks have been utilised.

2 -The Artificial Neuron.1 -“Network. The advantage of using a function like this is that it allows us to differentiate how close we came to the correct result. training by back-propagation involves the steps: • • • • • Present a set of training data Compare the network’s output to the desired output at each output neuron and calculate the error. based on the strength of the weights connecting them to the output neurons. This is far preferable to having to perform lengthy and expensive re-calibrations on any regular basis (Robot Control. 2.Back-propagation overview In essence.Back-propagation 2. allowing us to apply our learning algorithm. Assign responsibility to each of the neurons in the previous level. In order to use a back-propagation training algorithm you must have a non-linear activation function for your artificial neuron. This rule essentially describes adjusting the weight using the difference between the expected and actual outputs.cs” lines 130-133. Adjust the weight of the output neurons to lessen the error. the disadvantage of a differentiable function is that it is by the use of these functions that local optima become apparent in the search space. The sigmoid function is in the source code at 9.5.3. A commonly used function is the sigmoid function as described in section 2.5.2 . Adjust the weight of the neurons to minimise the responsibility. Page 20 of 225 .Back-propagation in depth In a single layer neural network (or with a single node) the perceptron rule is applicable. 2008).1 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 themselves.1. 2.1.5 .

as the effect that the change in the weight will have on the rest of the network is a missing factor. this rule has no positive effect in a multi-layer neural network.2 -The Artificial Neuron) an alteration can be made to the perceptron learning rule allowing this difficulty to be overcome.Altered perceptron learning rule (Matthews. ℎ : = [1 − ] . Back-propagation for the Uninitiated. 2002) In this equation “w” and “x” are as before. 2002) Therefore. such as the sigmoid function (see 2. = 1− Equation 4 . Back-propagation for the Uninitiated. Back-propagation for the Uninitiated. Back-propagation for the Uninitiated. 2002): ∆ = ℎ = − Equation 5 . Back-propagation for the Uninitiated.Altered delta calculation rule for the output layer (Matthews. 2002) Calculating the hidden layer deltas is a little more complicated. The other two values “di and yi” are the actual and desired outputs. “n” is the learning rate of the neural network. Provided you are using a differentiable function.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 ∆ = ℎ = − . the following alteration to the perceptron learning rule would serve our purposes (Matthews. respectively. Assuming the sigmoid activation function is in use. Equation 3 – The delta rule (Matthews.differentiated sigmoid function (Matthews. The effect that the change will have on the following neurons in the network is an extra factor. 2002) However. The sigmoid function differentiates very neatly. Page 21 of 225 .1. the output layer nodes training function can be written as: = 1− − Equation 6 .

to calculate the delta for a hidden layer requires the deltas for the following layers.Genetic Algorithms Genetic algorithms attempt to copy natural laws to a certain extent in order to apply a random search pattern to a defined search space.6 .The FANN neural network library (Nissen. although as it turned out I did not). Most organisms evolve by means of sexual reproduction and natural selection (Darwin. Back-propagation for the Uninitiated. in a language that can be linked to C# fairly easily and it also supports cascade neural networks. in order to apply this learning mechanism you start at the output of the network and work your way towards the inputs. The FANN library was decided upon for several reasons. 2002) As demonstrated above. it is possible to “breed” solutions to problems. 1859) Page 22 of 225 . against which to compare the genetic algorithm trained neural network.3 -Libraries Researched) for other libraries investigated) implements a back-propagation solution and this library will be used in the project to implement a feed-forward back-propagation neural network.Altered delta calculation rule for the hidden layer(s) (Matthews. By using similar techniques to natural selection.3 .Back-propagation library 2. Therefore. 1859) very neatly avoids one of the larger problems involved in software design: how to specify in advance all the possible permutations of the original problem and how the program should react to those permutations. FANN) The FANN library (see (9.7 . saving me finding a separate library.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Equation 7 . 2. open-source (allowing me to alter it if I needed too.5. 2. easy to implement. it is fast. Natural selection (Darwin.

In 1965 one Ingo Rechenberg. swapping genetic material. built on papers both by himself and by other researchers. Mutation plays a part in this as well. In 1975. many researchers had independently developed evolution-inspired algorithms for function optimisation and machine learning but did not get much follow up to their work. this application was largely theoretical. then of the technical University of Berlin introduced a technique he named evolution strategy. by the author John Holland. this mixing leads to much faster evolution than if off-spring simply copied the genetic material as it is in the parents. the publication of the book “Adaptation in Natural and Artificial Systems” occurred. 2. mostly designed to help evolutionary biologists’ model aspects of natural evolution. Also in 1975. By 1962. but this variety helps to prevent stagnation in the gene pool. genetic Page 23 of 225 . to alter slightly in a random way.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Sexual reproduction ensures mixing and recombination of genes. This could result in either an effective or defective solution. although this technique involved no crossover. genetic algorithms were becoming more widely used.1 . including noisy and discontinuous data. occasionally causing the genetic information. a dissertation by Kenneth De Jong established the potential of genetic algorithms by showing that they perform well on a wide variety of data. however. selection and crossover. This book presents the concepts of adaptive digital systems using mutation. during this process. simulating processes of biological evolution. During the process chromosomes line up together and then cross over partway along their length. or indeed multiple genomes in a population.7. By the early 1980s.History of Genetic Algorithms Genetic algorithms first appeared on computers in the late 1950’s and early 1960’s. At first. across a range of subjects.

2 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 algorithms soon moved into commercial territory and nowadays help solve many problems in many different areas. (Marczyk. more conventional. with processing power exponentially dedicated to better areas as the exploration progresses. in the same way that the average response of a relatively small percentage of the population of any given country.2.Advantages and Disadvantages of Genetic Algorithms 2. or further up the hierarchy. proceeding with other more likely candidates. Many other algorithms are largely serial and can only explore a search-tree so far in any one direction before abandoning all progress and starting again from the beginning. methods. Thanks to the parallelism that is a genetic algorithms main advantage. Genetic algorithms effectively explore many different branches of the tree at once and when a certain branch turns out to be non-optimal.7. 2004) 2.Advantages The primary advantage of genetic algorithms is their inherent parallelism.7. abandon that search. The real advantage of parallelism however is that by evaluating the relatively small search-space that it does. a genetic algorithm is implicitly evaluating a much larger group of individuals. 2004).1 . can be used to fairly accurately predict national trends. they are very well suited as a means of exploring search spaces too large to be effectively searched in a reasonable amount of time by other. This leads to the exploration of a large proportion of the search space. These are “non-linear” Page 24 of 225 . This is “Schema Theorem” and allows a genetic algorithm to move towards the searchspace with the most promising individuals in a timely fashion and then select the best member of that group (Marczyk.

However. 2004). In infinite or very large search spaces. or never contained in their search space in the first place. or merely very good local optima. Selection allows the pruning of the least promising searches. the random nature of the various starting points of the initial population and the other methods they employ. they produce random changes to their candidates and then use the objective function to determine whether those changes are positive or negative overall. Genetic algorithms have no prior knowledge about the problem they are trying to solve. due again to the parallelism of their approach. each component’s fitness is individual and any improvement to individual component's fitness is necessarily an improvement to the whole. Many real life problems are not like this. genetic algorithms tend to give good results compared to other search strategies. such as crossover and mutation. In a linear problem. crossover allows promising solutions to share their success and mutation allows random changes in the local search space of a given solution. One of the largest strengths of genetic algorithms can at first glance appear to be their largest weakness. This allows them to discover solutions that other algorithms may have over-looked. selection and mutation a genetic algorithm is metaphorically similar to a large number of parallel depth first algorithms.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 problems. it is hard to know whether we have reached the global optimum. each searching their own space for the best solution. they tend towards non-linearity where altering one component's fitness positively may make other components less fit and cause a ripple effect across the whole system. due to their strength in navigating large search spaces (Marczyk. A third advantage of genetic algorithms is that they do not tend to be easily trapped by local optima. making genetic algorithms an effective way to search them. A good example of this is the concept Page 25 of 225 . This non-linearity results in a huge increase in the search space. Without crossover.

For example in the diagram below. etc. with each number representing some distinct part of the solution (Marczyk. a comparatively smaller segment of the search space will be covered and the eventual solution may take longer to reach or never be reached as a result. A second disadvantage of genetic algorithms is that the fitness function is crucial to the development of a suitable solution for the problem. A third issue is setting the correct mutation rate. the genetic algorithm may end up solving an entirely different problem from the originally intended one. population size. If the mutation is too high. there is no slope leading to it. the investigation may cover too little of the search space to find the optimum solution (Marczyk. rediscovered by genetic algorithms. Generally. then genetic algorithms (and most other search techniques) are no better than a random search for finding the solution. where the actual code of the program forms the genomes in the genetic algorithm and it can swap portions in or out. 2. Likewise. 2004). 2004). Another method is to use genetic programming. if the size of the population is too low. integer. or adjust them. the global optima is at the far right. but denied a patent for several years because it ran counter to established beliefs (Marczyk. Page 26 of 225 . if the solutions to various problems are in areas of the search space that the fitness function would not consider likely.Disadvantages One of the disadvantages of a genetic algorithm is the necessity of representing the problem the genetic algorithm is trying to solve.2 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 of negative feedback. A further problem is that fitness functions can be deceptive.7. the system will never converge towards a suitable solution. 2004). If you have a poorly written fitness function. or real-valued numbers.2. however that is only one point. however if it is too low. genetic algorithms use strings of binary. therefore no gradual increase can lead you to it. in a form that the genetic algorithm can use.

There are various methods for solving this. 900 800 700 600 500 400 300 200 100 0 0 2 4 6 8 10 12 Figure 4 . where analytical solutions exist they should be take precedence over genetic algorithms. unless by random chance it discovered the point with the most optimum solution. 2004). 1996). In this example a properly constructed genetic algorithm would be likely to settle on the marginally less optimal.local optimum (units are undefined in this case). but altogether easier to find. which will then reproduce abundantly. lowering the diversity of the population and resulting in genetic algorithms possibly falling into the local optima that the mutation represents. Page 27 of 225 . including sigma scaling and Boltzmann selection (Mitchell.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 you either have the solution or you don’t. Finally. This is because analytical solutions usually produce more accurate results faster than a genetic algorithm (or any heuristic method) is capable of (Marczyk. local optima in the centre of the diagram. A further problem is premature convergence: this occurs when a mutation early in the process produces a large leap in the fitness of that particular genome.

such as adding a “momentum” to the algorithm. 2.1 .5 -Back-propagation) which is a gradient descent algorithm and as such can be easily trapped in local optima (Branke.Representation of a Neural Network within a Genetic Algrithm As a rule. since you are merely increasing the size of the genomes being worked with. however. to try to help it move beyond local optima.Training Neural Networks with Genetic Algorithms Training neural networks is a form of problem solving. involving finding the optimum values for a set of real numbers (connection weights) which will produce the least error in the results from the neural network.7 -Genetic Algorithms) are good at avoiding being stuck on local optima due to their searching several regions of search space simultaneously. 1995). 1992).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 2. not altering them besides that in any way. The error surfaces associated with these problems tend to be highly variable and contain many local optima. In addition. 2. to test the appropriateness of a given genome. however the method of choice for training neural networks is usually back propagation (see 2.8.2 . these are not a complete solution (Blum. a neural network representation within a genetic algorithm is a concatenation of all the weighs in the neural network. Genetic algorithms also have the benefit that they do not place any restrictions whatsoever on the architecture of the neural network. Page 28 of 225 .8. There are methods to try to avoid this problem when using back propagation. the only information genetic algorithms require to train a neural network is an objective function (as described in 2.7 -Genetic Algorithms).Determining Weight Values with Genetic Algorithms Genetic algorithms (as described in 2.8 .

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Crossover can disrupt genes far apart. Because of this it is helpful to keep functional units close together (I. side by side). 1995). 2. The problem at hand dictates the input and output layer structures.3 .Using Genetic Algorithms to Determine Neural Network Structure As well as using a genetic algorithm to evolve the weight set for a fixed-structure neural network. Standard genetic algorithms use binary strings. however representing a large number of real valued numbers as a binary string.8. outputs to inputs. place a neurons input and output weights. These are all connected. a nodes input weights and the nodes of a layer. one or two hidden layers of an indeterminate number of nodes and one output layer. Page 29 of 225 . can lead to very large strings (thousands of bits) (Branke. One of the most important decisions when deciding how to represent a neural network within a genome is whether to use binary strings or real number strings. to form a feed forward network.e. but is less likely to do so for genes close together. The average neural network has an input layer. it is also possible to use a genetic algorithm to evolve a structure for a neural network.

the neural network will never learn to solve the problem (or come close enough to suggest a solution). given a specific problem. this is not necessarily the most efficient form for a network for any given problem and the trial and error method is an inefficient method for determining the correct architecture for an optimally performing neural network for the problem at hand. On the other hand. if the neural network architecture is too complex. there is currently no method by which to establish. Although network architecture plays such a large role in network performance. Page 30 of 225 . the neural network will learn fast but will generate too specific a solution and will not generalize between similar inputs. a neural networks generalisation is its main strength and therefore you do not wish to lose it. However. There is also no method to check how optimal your solution is (Branke. As mentioned in section 1.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Figure 5 . If neural network architecture is too simplistic for the problem at hand.A feed-forward neural network This architecture is then altered (number of nodes in each hidden layer. 1995). a neural network topology to deal with that problem.0. number of hidden layers) by trial and error and intuition to establish a neural network that performs well for the problem at hand.

a representation is required which can accommodate all networks that will work for the problem but none that will not. an exception is possible. Back propagation learning rate is also a parameter for one of these areas. 1995). you must remove all connections to or from that node completely. thus meaning that in order to remove a node completely.Representing neural network structure Representing the structure of a neural network within a genetic algorithm is not as straightforward as representing the weight values of a neural network within a genetic algorithm. There are two methods for representing neural network topologies within a genetic algorithm. Ideally. whereas high-level encodings group together connections and/or nodes. the network is divided into areas and for each area the number of nodes and the density of connections to other nodes is determined (Branke. indirect. thus eliminating the searching of meaningless space and by definition. In direct encoding. or low-level encoding and the weak. In cases where a genetic algorithm determines the number of hidden layers and the number of nodes per hidden layer and the network itself is interconnected fully. Low-level encodings specify each connection individually.3. In one version of high level encoding of a neural network. the strong. the genetic algorithm represents the neural network by means of its connections. direct. Page 31 of 225 .1 .8. 1995). being able to contain the most optimum solution for the problem at hand. or high-level encoding (Branke.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 2.

will not ever succeed in learning the problem. by limiting the potential size of the networks evolved and preventing the generation of overly large and complex networks but on the other hand.2 neural networks that are too large will solve a problem fast but will not generalize well and therefore their usefulness is limited severely. These mapping methods are therefore more suitable for small networks or networks with a small number of connections.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 One of the reasons why alternatives to low-level encoding arose is the exponential rise in the number of connections. This could be useful. 1995).9 . and the moving target problem (Fahlman & Lebiere. Therefore. Page 32 of 225 .Cascade Correlation The cascade learning architecture is a method by which the training algorithm builds the neural network as it proceeds. it is impossible to know whether these large and complex networks may be the most suitable way to solve the problem at hand. The Cascade-Correlation Learning Architecture. This can lead to human bias entering the system and affecting the outcome. as the size of the neural network increases. which is obviously undesirable. The algorithm attempts to solve the issues associated with back-propagation that produce slow learning. This section largely drawn from information in: “Curved trajectory prediction using a self-organizing neural network” (Branke. As mentioned in section 5. Neural networks that are too small on the other hand. These are the Step-size problem. A problem with high level encoding is that regular networks are favoured but not every network is equally probable. 2. 1991). a balance between speed of learning and generalisation exhibited in the trained network is necessary.

with the aim of getting as close to a solution as possible.The definition of 's' given that 'o' is the network output at which the error is measured. After a number of training cycles pass with no substantial improvement (the precise number being determined by a user-supplied “patience” parameter).5. 1988). In order to maximise "s".2 -Back-propagation in depth) or the quick prop rule (Fahlman. The Cascade-Correlation Learning Architecture. plus the outputs of any previously created units. 1988). with a value of one. the candidate unit’s value. It then trains the adjustable weights using the perceptron rule (as described in 2. The output units may use either a linear or differentiable activation function. There is also a bias input. The “candidate” nodes take the neural net inputs. 1991). An Empirical Study of Learning Speed in Back-Propagation Networks. = −` −` . 'p' is the training pattern. Page 33 of 225 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 The system begins with a basic neural network of inputs and outputs (numbers of each dictated by the task as usual) and no hidden layer. 'v' is the candidate unit's value. The connection between each input and output is via an adjustable weight. the initiation of a number of “candidate” units takes place and trained using either the perceptron rule (2.5. or if the networks output is of sufficient quality the training is ended. 1991). and “Eo”. or sometimes the quick prop rule (Fahlman. either an extra node is added. the partial derivative of "s" with respect to each of the candidate unit’s incoming weights "wi" is a necessity. The Cascade-Correlation Learning Architecture. and attempt to maximise “s” the sum overall output units “o” of the magnitude of the correlation between “v”. If a trigger occurs for the addition of a unit.2 -Back-propagation in depth). An Empirical Study of Learning Speed in BackPropagation Networks. Equation 8 . the residual output error at “o” (Fahlman & Lebiere. the quantities ‘`v ’ and ‘`Eo’ are the values of v and Eo averaged over all patterns (Fahlman & Lebiere.

and the inputs for that node are frozen. where ‘oo’ is the sign of the correlation between the candidates value output ‘o’. After these computations are complete. Adjustable weights link the outputs for this node and all the output nodes inputs.The partial derivative of ‘s’ with respect to each of the candidates incoming weights ‘wi’. The training iterations for these adjustable weights in the main network then continue. The Cascade-Correlation Learning Architecture. Page 34 of 225 . Once the correlation on a candidate has reached an acceptable level. Expansion and differentiation of this calculation can then take place. a gradient descent to maximize "s" can take place. ‘f'p’ is the derivative for pattern 'p' of the candidate unit's activation function with respect to the sum of it’s inputs and ‘Ii. . −` ′ .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Equation 9 .The partial derivative of ’s’ (Fahlman & Lebiere. The connection of the best of the candidates to the neural network as it stands takes place. or the completed training epochs reaches a maximum level. Equation 10 . 1991).p’ is the input the candidate receives from unit ‘i’ for pattern ‘p’. = .

The initial state of a cascade correlation neural network. Figure 7 . 1991) Cascade-Correlation 1991). The square econd connections are locked weights (Fahlman & Lebiere. The circular connections are adjustable weights (Fahlman & Lebiere.William Sayers 05025397 Genetic Algorithms and Neural Networks Neural Milestone 3 Figure 6 . Page 35 of 225 .The second state of a cascade correlation neural network. The Cascade Correlation Learning Architecture. A first node has been added. The Cascade Correlation Learning Architecture. 1991) Cascade-Correlation 1991).

The third state of a cascade correlation neural network.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Figure 8 . The Cascade-Correlation Learning Architecture. with two nodes added (Fahlman & Lebiere. Page 36 of 225 . 1991).

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3

2.10 - C# User Interface Programming
C# uses the Winforms API to create user interfaces. It allows you to “draw” objects on a form in a visual manner to design your layout and then access and alter properties and subproperties of those objects in your programming code. A separate thread for each Window in Winforms exists, in order to facilitate performance on modern multi-core machines, although it does make it more difficult to move data between the forms. A method using “delegate functions” (similar to “function pointers” in C++) and “events” (Lysle, 2007) is the method I will use. Another challenge is making the neural network training independent of the user interface, in order to avoid the interface freezing whilst training a neural network. The solution I plan for this is to use the background worker object provided with C#, which allows you to assign a delegate function to be your background task, and will trigger events upon completion, for which event handlers will be executed (Microsoft, BackgroundWorker Class, 2009). Making a C# user interface interact with C++/CLI dll’s simply involves adding a reference to the dll in your project, via the Visual Studio® user interface.

2.11 - Writing C++/CLI Wrappers for Unmanaged C++ code
The cleanest method of connecting a native C++ or C dll or piece of code to a managed C# interface is via a C++/CLI bridge class. The VC++ compiled can swap between managed and unmanaged code on the fly whilst compiling and allows you to have both managed and unmanaged code (in C++) as part of the same project. Thus, a managed class can communicate with the unmanaged class (performing appropriate marshalling of the non-primitive data types (Microsoft, Using C++ Interop (Implicit PInvoke), 2009)) and the C# code can then access it directly.

Page 37 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3 Although it is possible for a C# program to interact with an un-managed dll directly, this method is less error-prone and more likely to be robust and stable, as well as making for cleaner code.

2.12 - Application of Research
The research presented here will help to produce the deliverables for milestone two and milestone three of my final year project: I plan to combine C# for the user interface and C++ for the neural network back-ends in order to create an efficient, powerful and fast solution. Section 2.11 -Writing C++/CLI Wrappers for Unmanaged C++ code, covers the combination of C# and C++ so I will not bother to cover it here, I plan to use the DLL method, with one DLL combining managed and unmanaged C++ linked into my C #user interface code. The user interface will then have methods to adjust parameters such as the genetic algorithms population size, number of generations, etc. In order to test and see what kind of genetic algorithm learns and develops fastest. It will also allow you to specify the number of back-propagation iterations. The use of this application will be in evaluating the difference in performance between a genetic algorithm, back-propagation algorithm and cascade correlation algorithm as neural network training algorithms.

3 - Design
The purpose of this application will be to allow the user to run a back-propagation trained neural network, or a genetic algorithm trained neural network and present the results of the network in such a fashion as to allow analysis and comparison of the training methods and their suitability in different situations.

Page 38 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3

3.1 - Program Requirements
The program must be able to: • • • • • • Run a genetic algorithm on a feed forward fully interconnected neural network to train the network. Run a back propagation algorithm on a similar network to train the network. Run a cascade algorithm on a similar network to train the network. Allow the user to set appropriate variables (see 3.4.2 -New Network Wizard) to adjust the execution of the learning algorithm selected for the neural network. Allow the user to select which learning algorithm they wish to use. Display the results of running that algorithm on a neural network in a meaningful fashion.

3.2 - Design of the Class Structure
Before I discovered a simple to program and effective design, I tried and discarded two designs. I eventually used the third design and it is that third design documented here. The two prior designs to this involved representing neural networks via objects and compositions of objects in vectors, which lead to flexible, but slow and overly complex code upon experimentation. The final solution, the solution documented here, is a far simpler one (in terms of the neural network representation, it is far more detailed in the other areas than the initial designs were). This solution loosely follows a model, view, controller architecture. The solution enables easy replacement of sections of code as long as the interfaces and outputs remain the same. For more information on the class structure, see section 9.2 -UML Class Diagram.

Page 39 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3

3.3 - Linking C# code to managed dll’s
Linking C# code and managed dll’s is a simple process (in Visual Studio) as explained in (2.11 -Writing C++/CLI Wrappers for Unmanaged C++ code) of selecting the references section in the C# project in question and adding the dll you wish to refer too. Then you can simply add a “using” directive to use the dll namespace and create, destroy, and use the classes within as if they were native C# classes. This can be seen in (9.1.2.6 “Algorithms\BackProp.cs”) and (9.1.2.4 -“Algorithms\CascadeCorrelation.cs”) with the “using FANN_Wrapper;” directive. This is superior to the direct C# to unmanaged code linking that used in milestone two, since it makes for cleaner and thus more easily maintainable code and the code is more robust in this fashion. The robustness comes from the fact that C++/CLI is far better at interacting with unmanaged C++ code than C# is. Indeed – they can even be in the same source file and compiled into one object. Therefore, by using a C++/CLI wrapper to interface to the unmanaged C++ and then interacting with the C++/CLI wrapper/s errors are less likely to arise.

3.4 - Design of the User Interface
The construction of the user interface utilised the Winforms based tools in visual studio 2008 (see 2.10 -C# User Interface Programming). I also used the ZedGraph control (ZedGraph) to display the graph of my results.

3.4.1 - Main Form
The main form is the crux of the application (as is default in C# Winforms programming) and if the main form is closed, all other forms and classes are disposed and the application terminates.

Page 40 of 225

Page 41 of 225 .Network and Output tab This tab needs a method by which to initialise the adding of a new network (3.1.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 3. One display will show network details and one display will show the training and testing output. Three buttons will therefore be present. which simply needs to display the graph of the mean squared error over the current epoch/generation. Figure 9 – Network and output tab design (Main form) 3.4.1.Mean Squared Error Graph tab The mean squared error graph tab is a simple tab.2 .4.1 .4.2 -New Network Wizard) a method to train the current network and a method to test that training. along with two displays.

4. to show the training data and to show the testing data.Mean squared error tab design (Main form) 3.1. Figure 11 .3 .Dataset tab design (Main form) Page 42 of 225 .Dataset tab This tab needs two display sections.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Figure 10 .

7 Genetic Algorithms.4. the default is genetic algorithm). out of three choices previously mentioned (2. Verification of all data takes place before the form passes control back to the main form and triggers a main form event with a parameter that takes a specially constructed class to transfer the data from the wizard.Learning Algorithm tab On the learning algorithm tab you select the learning algorithm for the neural network you are creating.2.4.2 .5 -Back-propagation.New Network Wizard The new network wizard is the form that appears when the user selects the creation of a new network.Design of the training algorithm tab (Wizard form) 3.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 3. Figure 12 .2. 3.2. It allows you to move in a logical progression through the set up process of a new neural network and associated training algorithm.1 -Learning Algorithm tab (if no algorithm has yet been selected.4. 2.Network Settings tab This tab allows you to set appropriate settings for the network and training algorithm you have selected in 3. 2.9 -Cascade Correlation).1 .4. Page 43 of 225 .2 . The user can progress through the steps via the “previous” and “next” buttons or via the tabs at the top in whatever order you choose.

Design of the network settings tab (Wizard form) 3.Dataset selection tab design (Wizard form) 3.3 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Figure 13 . or to choose custom data sets from your hard disk for training and testing data (appropriately formatted in the FANN style).4.Working Form The working form is merely a small display shown as a dialog (lock out access to displaying form) which shows whilst a neural network is training.2.3 . Page 44 of 225 .Dataset tab The dataset tab allows you either to select from two pre-setup datasets. Figure 14 .4. both to protect the main form from data inputs that may cause training problems and to provide a visual indicator of work taking place.

1 .4 . one tab with XOR information and one tab with Fishers Iris Data information.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Figure 15 .4. project description).About Form The about form displays information about the application (author. Figure 16 . Clicking the datasets button or the licenses button displays the appropriate information. displays licensing information and displays information on the pre selected datasets. Page 45 of 225 .About form design (About form) 3.4.Datasets display Clicking the data sets button displays a tabbed interface.Working form design (Working form) 3. project supervisor. clicking the same button again hides that information.4.

Page 46 of 225 .2 .Implementation The implementation is largely in C#.Licenses display Clicking the licenses button displays a small information panel with selection methods. FANN) library written in C and linked as a static library to the managed FANN_Wrapper dynamic link library.4.Licenses display design (About form) 4 . Figure 18 . with small portions in C++/CLI and the back-end FANN (Nissen.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Figure 17 . depending on your selection.Dataset display design (About form) 3. it will display the appropriate license.4.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 4.cs”.1.1 . mostly via the visual design tools found in visual studio 2008.1.1.Network and Output tab Figure 19 .2 -“FrmMain.Main Form The main forms code is in section 9.1.1. 4.1 .User Interface Implementation The implementation of the user interface is mainly in C#.Network and Output tab Page 47 of 225 . 4.1 .

Mean Squared Error Graph tab Figure 20 .Mean squared error graph tab Page 48 of 225 .1.1.2 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 4.

The dataset tab Page 49 of 225 .Dataset tab Figure 21 .1.1.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 4.3 .

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3

4.1.2 - New Neural Network Wizard
4.1.2.1 - Learning Algorithm tab

Figure 22 - Learning Algorithm tab

Page 50 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3

4.1.2.2 - Network Settings tab

Figure 23 - Network Settings tab

4.1.2.3 - Dataset tab

Figure 24 - Dataset tab

Page 51 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3

4.1.3 - About Form

Figure 25 - About form, main view

Page 52 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3

4.1.3.1 - Datasets View

Figure 26 - About form, Datasets view

Page 53 of 225

Page 54 of 225 .Working form 4.Working Form Figure 28 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 4.10 -C# User Interface Programming) via events and delegate functions.1.About form.3.1.2 .5 . Licenses view 4.1.Licenses View Figure 27 .4 .Passing Information between Forms Passing the data between the wizard form and the main form takes place as previously described (2.

1. Page 55 of 225 .1. 4.1.2 -“FrmMain.2 -“FrmMain.6 .cs” lines 120-129).1.1.2 -“FrmMain. lines 29-44).cs”. lines 306-353.1. BackgroundWorker Class.cs”. This leads to a user interface that is much more accessible (Microsoft. are associated with a delegate function to respond to that event (Lysle.1. The trigger of a new event takes place in the Wizard form when you are on the last tab and the user clicks “next” (see 9.1. The declaration and association of a delegate and an event takes place in the form that needs to generate an event (9. the processing of the class sent from the Wizard form takes place (see 9. The two-event handling functions are at 9.1. 2009).cs” lines 70-119). Firstly. (See 9.1.1.3 -“Wiz1. Events. 2007). lines 29 – 31). The “NetworkUpdatedEventArgs” class is a storage class for passing data (see 9.1. when instantiated.cs” lines 145-163). the background worker class is instantiated and supplied with the event handler references functions (see 9.Keeping the User Interface Active whilst Processing is Occurring The user interface delegates the processing to a background worker class. lines 266-281). Then the runworkerasync function executes and then shows the working form.1.cs”.1.cs”.1. In the event handler in the main form.3 -“Wiz1.2 “FrmMain.1.1.2 -“FrmMain.1 -“NetworkUpdatedEventHandler”).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 An event is a method of notifying a listening object that something has occurred. The declaration of a new event handler in the main form takes place prior to displaying the new neural network wizard form (see 9.

1 . • • The sorting implementation in GANetwork.2.Back-propagation and Cascade Correlation Implementation The FANN library implements both cascade correlation and back propagation (Nissen.1.cs” lines 176204). and didn’t suffer from the memory leaks (being managed code) so that’s the version that exists in the application today (see 9. The neural network specific functionality added to the network model in the network specialisation class GANetwork (see 9. when the C++ implementation of genetic algorithms that I wrote functioned.3 Page 56 of 225 . The compilation of the FANN library is static.1.5 “Algorithms\GeneticAlgorithm.5 -“Algorithms\GeneticAlgorithm.Genetic Algorithm Implementation Although the original design was to implement genetic algorithms in C++. Once this implementation was complete.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 4. after having used C# extensively in the UI I felt more confident about the performance loss and the use of the programming language itself.cs” lines 51-126) 4. I decided to try re-implementing it in C#. Therefore.3.cs” lines 32-44).cs that links into the . There is an unmanaged C++ wrapper (9.2 -“GANetwork. but had issues with memory leaks and was slow.2.cs” for the C# implementation) The main points to note in the code as it stands are: • The crossover (single-point) and mutation (random increment or decrement) implementations (see 9.3.1.1.2.2 . FANN) which is a C programming language based neural network library. and the Dll linked to this to allow access to the functionality.1.NET built in functionality (see 9.4.2 -“GANetwork.NET sort methods in the sorting part of the genetic algorithm). it was both faster than the C++ implementation (due mostly to my use of build in .

1.cpp”.h”.1 -“Network.1. 9.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 “UnManaged_FANN_BackProp.2.1 -“FANN_Wrapper. the separation of the textual inputs takes place. which allows it to manipulate the FANN functions in the static library through this class. In the initialisation routines (lines 28-124). 9. number of inputs. number of outputs) separates the data into two-dimensional arrays suitable to be passed to other functions.2 -“DatasetParser.4. The inputs and weights are structures so that the coordinates for an input.cpp”) for the unmanaged C++ classes. Three jagged arrays are utilised.4.h”. which store respectively.4.3.4 . via a number of designated separator characters. layers.4.5 -“UnManaged_FANN_CasC. From there. the inputs. 9.4 “UnManaged_FANN_BackProp. followed an object-based structure (network. 9.cpp”) for the Cbased FANN functions in the DLL and a C++/CLI wrapper (9. which using the data from the heading (number of epochs of data.2 -“FANN_Wrapper.cs”). into arrays of strings. In the constructor (lines 41 – 77). these arrays form parameters for the other functions. 4.cs”) parses data directly from the rich text boxes that they’re displayed upon on the main form.Data Parser Implementation The data parser (9. the weights.h”. and the outputs. the initialisation of all the arrays to their correct sizes takes place and then the initialisation of the weights to a random double between zero and one.1.1. and nodes) the eventual implementation turned out to be basic yet effective (9.4.6 -“UnManaged_FANN_CasC. will also be the coordinates for its corresponding weight (lines 17-19). 4. C# accesses the C++/CLI class directly. Page 57 of 225 .3 . via references to these objects passed down via the hierarchy (references are passed to avoid the necessity of passing entire strings down the hierarchy).1.Neural Network Implementation Although two previous attempts made at programming a neural network.1.1.

to run the output layer. so repetition is instead used.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 The main functionality of the neural network class can be found in the “internal” region (lines 126-235). I tested the neural network implementation by creating a neural network with structure 2. and creating a neural network with a similar structure on an excel spreadsheet.4 -Datasets) (ten times for XOR. Then entered the weights from the neural network program (by means of using the visual studio debugger to check variable values) into the spreadsheet and checked that the outputs coincided.1 . 5. the dataset is too small to split. 1. The target error was 0. 5 . All nodes determine their outputs using the sigmoid function (lines 130-133).02 for Fishers iris data and the virus classification data. Page 58 of 225 . and hidden layer which are all called by the public “run()” function (lines 241-270). recording took place for analysis. The genetic algorithm and back-propagation neural networks each had four hidden nodes for each dataset. As the program presented the data in each case. 3. five times for Virus classification and Fishers iris data).01 for XOR and 0. input layer.Testing Data Testing took place several times for each algorithm and dataset (see 9.XOR In the XOR dataset. there are three functions.

Genetic Algorithm testing data (XOR) 5.009829335 0.00999799 0.00990403 0.009157809 Table 2 .00995716 0.009157809 Testing MSqE (GA) 0.009248076 0.009162436 0.00998959 0.1.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 5.00987211 0.00957518 0.006150786 0.0099765 0.00998032 0.009162436 0.009375132 0. because of floating point inaccuracies converting between double precision floating-point numbers in my code and single precision in the FANN code.00988649 0.1.00988863 0.009969323 Testing MSqE (BP) 0.00990161 0.009829335 0.009375132 0.009885499 Table 3 . Iterations (BP) 345 287 319 327 301 323 421 315 362 325 332.00991907 0.00881867 0.009248076 0. the mean squared error from the network result and the testing mean squared error are different despite the data being the same.Back-propagation In the back-propagation testing data.009888967 0.009962559 0.0099113 0.00881867 0.Back-propagation testing data (XOR) Page 59 of 225 .00983181 0.006150786 0.009566951 0.00989974 0.00992081 0.009962559 0.Genetic Algorithms Iterations (GA) 591 407 500 466 499 534 675 557 357 939 552.00992291 0.009888967 0.1 .009566951 0.00998027 0.00997699 0.2 .5 Result MSqE (BP) 0.00957518 0.0098402 0.00999069 0.5 Result MSqE (GA) 0.

78E-14 1. At each node addition.14E-14 2. but in the case of cascade training.Cascade Correlation In the cascade correlation data.3 .00798691 0.00889787 0. Nodes (CC) 2 2 2 2 2 2 2 2 2 2 2 Result MSqE (CC) 0.00585159 0.00573252 0.Cascade Correlation testing data (XOR) Page 60 of 225 .83E-13 2.1.18E-13 7.32E-13 2.008039565 Testing MSqE (CC) 7.00960884 0. the call back function triggers.00855044 0.02E-13 1.99E-13 1. the network is trained more after that point (output weight training).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 5.00863005 0. the testing result differs quite widely from the network result.02E-13 2.22E-13 1.00992534 0. which is when the program records the mean squared error.00656719 0.86E-14 1.0086449 0.43545E-13 Table 4 .

01 0.Graphs 1000 900 800 700 600 500 400 300 200 100 0 0 5 10 15 Iterations (GA) Iterations (BP) Nodes (CC) Figure 29 .1.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 5.Number of iterations to achieve target (XOR) 0.002 0 0 5 10 15 Result MSqE (GA) Result MSqE (BP) Result MSqE (CC) Figure 30 .004 0.008 0.Mean squared error upon last training test (XOR) Page 61 of 225 .006 0.012 0.4 .

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 0.008 0.002 0 0 5 10 15 Testing MSqE (GA) Testing MSqE (BP) Testing MSqE (CC) Figure 31 .012 0.The mean squared error at testing (XOR) Page 62 of 225 .01 0.006 0.004 0.

01956255614 Testing mean squared error (GA) 0.2 .0139982 0.Cascade correlation testing data (Iris data) Page 63 of 225 .019763877 0.Genetic algorithm testing data (Iris data) 5.0193595 0.0158679 0.014144 0.0199918 0.1 .019763877 0.018789438 0.2.019775896 0.019719802 0.0199999 0.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 5.0199719 0.019719802 0.0189456 0.Back-propagation testing data (Iris data) 5.018789438 0.Fishers Iris data 5.0164964 0.019751257 0.01956255614 Table 5 .0141406 0.0196297 0.01998996 Testing mean squared error (BP) 0.3 .014116 0.0143778 0.0199916 0.0160925 Table 7 .0141935 Table 6 .2.0198416 0.0199892 0.019775896 0.0143806 0.Cascade Correlation Nodes (CC) 2 2 2 2 2 2 Mean squared error (CC) 0.01950238 Testing mean squared error (CC) 0.Back-propagation Iterations (BP) 434 529 474 391 527 471 Mean squared error (BP) 0.2.2 .Genetic Algorithms Iterations (GA) 1928 1834 1277 2835 8716 3318 Mean squared error (GA) 0.019751257 0.0195853 0.0199972 0.0140669 0.

02 0.019 0.Graphs 10000 9000 8000 7000 6000 5000 4000 3000 2000 1000 0 0 2 4 6 Iterations (GA) Iterations (BP) Nodes (CC) Figure 32 .0186 0 2 4 6 Mean squared error (GA) Mean squared error (BP) Mean squared error (CC) Figure 33.0196 0.2.4 .0202 0.0188 0.0198 0.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 5.Number of iterations to achieve target (Iris data) 0.0194 0.Mean squared error upon last training test (Iris data) Page 64 of 225 .0192 0.

02 Testing mean squared error (GA) Testing mean squared error (BP) Testing mean squared error (CC) 0. tomato. and others.The mean squared error at testing (Iris data) 5.01 0. There are 18 measurements on each virus.025 0. The dataset is in order: • • • • Hordeviruses x3 Tobraviruses x6 Tobamoviruses x39 Furoviruses x13 Page 65 of 225 . which are the number of amino acid residues per molecule of coat protein.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 0.Virus Classification This dataset covers 61 viruses affecting several crops including tobacco.005 0 0 2 4 6 Figure 34.015 0. cucumber.3 .

Genetic Algorithms Iterations (GA) 1818 1421 896 11676 1068 3375.2 .019927361 0.017723052 0.0105845 0.0350318 0.3.0199916 0.0195173 0.0347797 0.0111416 0.Back-propagation Iterations (BP) 3275 10922 2800 3744 6295 5407.Cascade Correlation testing data (Viruses) Page 66 of 225 .019743 0.3.00979968 0.0190612094 Mean Squared Error testing (GA) 0.0194383 0.0190612094 Table 8 .1 .0165979 0.Cascade Correlation Nodes (CC) 1 2 1 2 1 1.Genetic Algorithms testing data (Viruses) 5.019092293 0.017043 0.3 .019092293 0.00980208 0.019927361 0.00640291 0.019861371 0.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 5.2 Mean Squared Error training (BP) 0.03833646 Table 9 .01870197 0.8 Mean Squared Error training (GA) 0.01870197 0.Back-propagation testing data (Viruses) 5.01983722 Mean Squared Error testing (BP) 0.4 Mean Squared Error training (CC) 0.0434423 0.011585152 Table 10 .0391179 0.0196959 0.0199621 0.0197935 0.0393106 0.019861371 0.3.016183342 Mean Squared Error testing (CC) 0.0185152 0.017723052 0.

3.Mean squared error upon last training test (Viruses) Page 67 of 225 .015 0.01 0.4 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 5.Graphs 14000 12000 10000 8000 6000 4000 2000 0 0 2 4 6 Iterations (GA) Iterations (BP) Nodes (CC) Figure 35 .025 0.005 0 0 2 4 6 Figure 36 .Number of iterations to achieve target (Viruses) 0.02 Mean Squared Error training (GA) Mean Squared Error training (BP) Mean Squared Error training (CC) 0.

versus backpropagations 333 iterations on average. genetic algorithms and back-propagation offer similar performance in solving the problem.Comparisons 6.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 0.Back-propagation and Genetic Algorithms 6.045 0.03 0.025 0. Page 68 of 225 . or possible inefficiencies in my programming.1.05 0.01.005 0 0 2 4 6 Mean Squared Error testing (GA) Mean Squared Error testing (BP) Mean Squared Error testing (CC) Figure 37 .015 0.04 0.Mean squared error at testing (Viruses) 6 .02 0.XOR As regards to the XOR problem. Genetic algorithms take around five hundred iterations on average to produce a workable solution when aiming for a mean squared error of below 0. Genetic algorithm solutions are slightly more accurate.035 0.1 .01 0.1 . but to a negligible degree in this scenario. The back-propagation implementation seems to be slightly faster in real-time but this is possibly attributable to managed code inefficiencies versus unmanaged code.

on this data set cascade training came to a satisfactory solution faster and more accurately than genetic algorithms. At a maximum of 150 epochs per node added (the default (Nissen. The cascade algorithm also achieves a higher degree of accuracy.2.XOR This problem was solved more effectively by the Cascade algorithm than by genetic algorithms.Fishers Iris Data Genetic algorithms clearly found this a difficult problem to solve as opposed to backpropagation. The genetic algorithm solutions also held their effectiveness through to the testing data with extreme consistency. 6.1 . FANN)) this is around 300 epochs to solve the problem.Cascade Correlation and Genetic Algorithms 6.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 6.2 . With an average of 3318 generations for genetic algorithms versus 471 epochs for back-propagation this became clear.3 . Page 69 of 225 .2 .1.8 generations versus 5407. The back-propagation generated networks were also more accurate overall (in fact on this particular problem analysed the testing data more effectively than the training data). 6.Fishers Iris Data Again. with an average of 3375. whereas the back-propagation solutions lost effectiveness on the training data.Virus Classification The solution of this problem was more effective by genetic algorithms than by backpropagation.2 epochs for backpropagation.2.2 . whilst the cascade algorithm used 2 nodes to solve the problems. genetic algorithms taking 550 generations on average to solve the problem to a suitable accuracy. 6.1.

7 . the unavoidable conclusion is that cascade training is the most effective of the three training methods examined. 6. Page 70 of 225 .Evaluation Overall.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 6. however.3 .Virus Classification With this data set cascade training again came to a satisfactory solution faster and more accurately than genetic algorithms did on the same data. They are roughly as effective as back-propagation training and although back-propagation is faster for the most part. solutions developed using genetic algorithms tend to hold their effectiveness through to the data set effectively. I feel the project has been a success. With more time. with the information presented and discussed here.Testing conclusions Genetic algorithms are a valuable and effective means of training neural networks. Compared to cascade correlation however. more in-depth testing using more varied datasets a clearer comparison and possibly drawbacks associated with cascade training may become evident. genetic algorithms fall down as a solution. although there are further investigations I would like to continue with given more time (and may continue with in my own time).3 .2.

1 . Research current information on training neural networks using genetic algorithms. and in which cases these methods are suitable. and in which cases these methods are suitable. Build an application that is capable of training a neural network with a genetic algorithm and with a backpropagation system. as well as previous research. leading to a more solid conclusion and more information on the various strengths and weaknesses of the various learning algorithms examined. and that allows manual adjustment of relevant variables. Using above application. Identify the most effective ways of training neural networks using genetic algorithms.” With more time more advanced tests could be undertaken using larger data sets.5 -Objectives): Research the fields of genetic algorithms and neural networks. Page 71 of 225 .How many of the original objectives were achieved? The original objectives were as follows (9. The only objective that I feel could be improved upon is the final objective. Build a user-interface for the aforementioned application allowing for easy alteration of appropriate variables. “Identify the most effective ways of training neural networks using genetic algorithms.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 7. evaluate training methods for neural networks.

These could be resolved fairly simply (I believe) by implementing the iDispose interface in my managed code to allow it to be destroyed upon demand instead of waiting for the garbage collector. Testing both with and without scaling and utilising different scaling methods.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 8 . Due to small conflicts between managed and unmanaged memory. the program occasionally causes an exception (most specifically when creating cascade or back-propagation networks in quick succession). Testing with datasets that would take large amounts of time to train. Improve memory management. Page 72 of 225 .Possible Improvements More advanced testing would be the biggest improvement that more time would make possible. Some forms of testing that would be possible are: • • • • • Testing with datasets with larger numbers of inputs Testing with large datasets that attempt to train the network to distinguish between small differences in data.

• • Attempt to improve the speed of the genetic algorithm.1 . for the reasons stated above. 8. Having the back-propagation algorithm in C# similar to the genetic algorithm would also allow a more “apples to apples” comparison. • Implement the cascade training algorithm in C# instead of C++. Making the forms shown as dialogs. Contrary to my opinion before undertaking extensive usage of C# I now believe the reduced development time. Page 73 of 225 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 8. but still minimize instead of blocking other computer programs. is worth the possibility of reduced performance.Possible Training Algorithm Improvements With more time. Implement the back-propagation algorithm in C# instead of C++.Possible User Interface Improvements The main improvements to the user interface that I would make if I had more time are: • • Increased number of variable options presented to the user (with appropriate defaults set). (the new network wizard and the working form) lock out the rest of the program. allowing more to be undertaken successfully. experimenting with multiple point crossover and other crossover forms. as well as implementing roulette wheel selection. I would experiment with the following training algorithm improvements: • Improve the genetic algorithms crossover and mutation algorithms.2 .

Although there was another section of the program compiled from source (the FANN library). This section also contains only the final draft of the code.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9 . and neither is the code for the prototype constructed in milestone two. The earlier attempts at genetic algorithm and neural network implementations are not included.1 . which this code is partially based upon.Source code The source code in this section is organised first by which part of the application it represents and then by what source file it originally resided in. I have not included it in the source code listing.Appendices 9. The code contains line numbering for pinpointing the precise lines in question within the body-text when crossreferencing. so that this listing contains only code I have personally written for the project. Page 74 of 225 .

View 9. System.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9.1.Classes { public class NetworkUpdateEventArgs : System.1 .Text. //Dataset identifier private int _TrDataSet.EventArgs { #region data /********************\ |* DATA *| \********************/ //Training Algorithm Identifier private int trainingalg.Controller. private int _CC_Reports. //Cascade Correlation private double _CC_Terr.1.1.Generic. private bool _CC_ITerr. private int _CC_MaxNeurons. private double _CC_LearningRate. System.Collections. System. namespace _05025397. Page 75 of 225 .“NetworkUpdatedEventHandler” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 using using using using System.Linq.1 .

private int _GA_HiddenL. int CCMaxNeurons. private int _GA_GenLimit. private int _BP_Reports. private int _BP_HiddenL. private int _GA_Mutation. int GAPopSize. private int _GA_Crossover.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000027 00000028 00000029 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 //Genetic Algorithm private int _GA_PopSize. private double _BP_LearningRate. private double _GA_Terr. private bool _BP_ITerr. Page 76 of 225 . private int _GA_Reports. bool CCITerr. private bool _GA_ITerr. #endregion #region constructor /********************\ |* CONSTRUCTOR *| \********************/ //Constructor to accept all necessary data public NetworkUpdateEventArgs (int Algorithm. //BackProp private int _BP_EpochLimit. double CCTerr. double CCLearningRate. int CCReports. private double _BP_Terr.

double BPLearningRate. _CC_Reports = CCReports. double BPTerr. int GAHiddenL. _GA_Crossover = GACrossOver. _GA_Terr = GATerr. int GAMutation. int BPReports. int BPHiddenL. _CC_MaxNeurons = CCMaxNeurons. _CC_LearningRate = CCLearningRate.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000057 00000058 00000059 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 int GAGenLimit. bool BPITerr. _CC_ITerr = CCITerr. _GA_GenLimit = GAGenLimit. //Backpropagation Algorithm _BP_EpochLimit = BPEpochLimit. int GAReports. int BPEpochLimit. _GA_Mutation = GAMutation. int GACrossOver. double GATerr. Page 77 of 225 . _GA_HiddenL = GAHiddenL. bool GAITerr. _GA_ITerr = GAITerr. //Genetic Algorithm _GA_PopSize = GAPopSize. _GA_Reports = GAReports. int TrData_Set) { //Algorithm trainingalg = Algorithm. //Cascade Correlation _CC_Terr = CCTerr.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000087 00000088 00000089 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 _BP_HiddenL = BPHiddenL. } } public bool CC_ITerr { get { return _CC_ITerr. } } public int CC_Reports { get { return _CC_Reports. _BP_Terr = BPTerr. } #endregion #region getters/setters #region cascade_correlation //Cascade Correlation public double CC_Terr { get { return _CC_Terr. } } #endregion #region genetic_algorithms //Genetic Algorithm public int GA_PopSize { get { return _GA_PopSize. _TrDataSet = TrData_Set. } } public int GA_GenLimit Page 78 of 225 . } } public int CC_MaxNeurons { get { return _CC_MaxNeurons. _BP_ITerr = BPITerr. _BP_LearningRate = BPLearningRate. _BP_Reports = BPReports. } } public double CC_LearningRate { get { return _CC_LearningRate.

} } public double BP_Terr { get { return _BP_Terr. } } public int BP_HiddenL { get { return _BP_HiddenL. } } public double BP_LearningRate { get { return _BP_LearningRate. } } public int GA_HiddenL { get { return _GA_HiddenL. } } #endregion Page 79 of 225 . } } public double GA_Terr { get { return _GA_Terr. } } public int GA_Reports { get { return _GA_Reports. } } #endregion #region back_propagation //Back propagation public int BP_EpochLimit { get { return _BP_EpochLimit.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000117 00000118 00000119 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 00000146 { get { return _GA_GenLimit. } } public bool GA_ITerr { get { return _GA_ITerr. } } public int GA_Mutation { get { return _GA_Mutation. } } public int BP_Reports { get { return _BP_Reports. } } public int GA_Crossover { get { return _GA_Crossover. } } public bool BP_ITerr { get { return _BP_ITerr.

System.Generic.1. System.Windows. System.“FrmMain.1.Controller { public partial class FrmMain : Form Page 80 of 225 .IO. } } public int TrainingAlgorithm { get { return trainingalg. } } #endregion } 9.Forms. namespace _05025397.Drawing. System.2 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000147 00000148 00000149 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 } #endregion #region external //General public int TrDataSet { get { return _TrDataSet. System.ComponentModel. System. System.cs” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 using using using using using using using using using System. using _05025397. System.Linq.Data.Collections.Text. using ZedGraph.

//Working dialog Controller.Working(). //Background thread static BackgroundWorker bw = new BackgroundWorker().DoWork += new DoWorkEventHandler(bw_DoWork). } #endregion #region internal_functions /********************\ Page 81 of 225 .Working workingdialog = new Controller.TrainingAlgorithm network.Resources. bw.RunWorkerCompleted += new RunWorkerCompletedEventHandler(bw_RunWorkerCompleted). #endregion #region constructor /********************\ |* CONSTRUCTOR *| \********************/ public FrmMain() { InitializeComponent().Icon = Properties.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 { #region data /********************\ |* DATA *| \********************/ //The neuralnetwork Controller. this.icon. bw.Algorithms.

MSEpane. this.zedGraphControl1.NetworkUpdateEventArgs e) { LoadDataset(e.Clear(). } private void updatenetwork(object sender.Legend.YAxis. SymbolType.XAxis.IsVisible = false. try { if (e.Title. Classes. MSE.Red.Text = "Iteration".Title. // Hide the legend MSEpane.AddCurve("".GraphPane. MSEpane.CurveList. MSEpane. //Titles MSEpane. GraphPane MSEpane = this.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 |* INTERNAL *| \********************/ private void plotgraphsMSE(PointPairList MSE) { this.zedGraphControl1.TrDataSet).Title.TrainingAlgorithm == 0) { Page 82 of 225 .Circle).Text = "MSE".Text = "Mean Squared Error".AxisChange().GraphPane. Color.zedGraphControl1.

e.Algorithms. } } catch { Page 83 of 225 .BP_Terr.GA_Mutation.BP_LearningRate.GeneticAlgorithm (e. } if (e. e.BP_EpochLimit.TrainingAlgorithm == 1) { network = new Controller. txtTrainData. e. e.BP_Reports.CC_Reports. e. e. e.Algorithms. txtTrainData.CC_ITerr.GA_Reports. e.BackProp (e.BP_HiddenL.BP_ITerr. e.GA_PopSize. e.GA_ITerr.CC_Terr. e. e. e. txtTestData). e. txtTestData). txtTrainData.TrainingAlgorithm == 2) { network = new Controller.CC_LearningRate. e. txtTestData).CascadeCorrelation (e.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 network = new Controller.CC_MaxNeurons.GA_GenLimit. e.GA_Terr.GA_HiddenL.GA_Crossover. } if (e.Algorithms.

OK. wizard. MessageBoxIcon.Wiz1.Text = network.Resources. Page 84 of 225 .XOR: txtTrainData. } private void LoadDataset(int Data_set) { switch (Data_set) { case (int)Dataset. "Error".Exit(). txtTestData.Text = Properties.Wiz1().NetworkUpdateHandler(updatenetwork).XORtest.NetworkUpdated += new Controller. MessageBoxButtons.Error).Show("Error creating new network". } } private void new_network() { Controller. } if (network != null) { txtNetSettings.Resources. Application.ShowDialog().network_details().William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 00000117 00000118 00000119 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 MessageBox.Text = Properties. wizard.XORtrain.Wiz1 wizard = new Controller.

string test = "". "Error!".OK.Resources.\\". case (int)Dataset. case (int)Dataset. MessageBoxIcon. } catch { MessageBox. "Select Training Data").FISHER: txtTrainData. try { train = LoadTxtFile(trainpath). break.FISHERtrain. string train = "".Text = Properties.\\".Resources. } } private void LoadCustomDataSet() { string trainpath = SelectTextFile(". "Select Testing Data").Show("Error: Problem loading training data".William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 00000146 00000147 00000148 00000149 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 break. MessageBoxButtons.Error). break.CUSTOM: LoadCustomDataSet().Text = Properties.FISHERtest. string testpath = SelectTextFile(". txtTestData. } Page 85 of 225 .

} catch { MessageBox.Text = test. data = Reader. } txtTestData. MessageBoxIcon. "Error!". StreamReader Reader = null. FileAccess.ReadToEnd().Error).Text = train.Show("Error reading selected file: " + path. Reader = new StreamReader(File).Show("Error: Problem loading testing data".Open.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000168 00000169 00000170 00000171 00000172 00000173 00000174 00000175 00000176 00000177 00000178 00000179 00000180 00000181 00000182 00000183 00000184 00000185 00000186 00000187 00000188 00000189 00000190 00000191 00000192 00000193 00000194 00000195 00000196 00000197 try { test = LoadTxtFile(testpath). MessageBoxIcon. FileMode.OK. txtTrainData. } catch { MessageBox.OK. } private string LoadTxtFile(string path) { string data = "". FileStream File = null.Read). MessageBoxButtons. "Error!".Error). MessageBoxButtons. try { File = new FileStream(path. Page 86 of 225 .

Close(). dialog.OK) ? dialog. dialog. if (File != null) File.FileName : null.*". } private string SelectTextFile(string initialDirectory. } #endregion #region menu_items /********************\ |* MENU ITEMS *| \********************/ Page 87 of 225 .Filter = "txt files (*.InitialDirectory = initialDirectory. string title) { OpenFileDialog dialog = new OpenFileDialog(). return (dialog. } return data.*)|*.Close().William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000198 00000199 00000200 00000201 00000202 00000203 00000204 00000205 00000206 00000207 00000208 00000209 00000210 00000211 00000212 00000213 00000214 00000215 00000216 00000217 00000218 00000219 00000220 00000221 00000222 00000223 00000224 00000225 00000226 00000227 data = "".Title = title. dialog.txt)|*. } finally { if (Reader != null) Reader.ShowDialog() == DialogResult.txt|All files (*.

EventArgs e) { this.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000228 00000229 00000230 00000231 00000232 00000233 00000234 00000235 00000236 00000237 00000238 00000239 00000240 00000241 00000242 00000243 00000244 00000245 00000246 00000247 00000248 00000249 00000250 00000251 00000252 00000253 00000254 00000255 00000256 00000257 private void newNetworkToolStripMenuItem_Click(object sender. } private void runToolStripMenuItem_Click(object sender. } private void exitToolStripMenuItem_Click(object sender.Show().Trained != false) { network. EventArgs e) { FrmAbout About = new FrmAbout(). EventArgs e) { if (network != null) { if (network.Test(). } #endregion #region buttons Page 88 of 225 . EventArgs e) { new_network(). } } } private void aboutToolStripMenuItem1_Click(object sender. About.Close().

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000258 00000259 00000260 00000261 00000262 00000263 00000264 00000265 00000266 00000267 00000268 00000269 00000270 00000271 00000272 00000273 00000274 00000275 00000276 00000277 00000278 00000279 00000280 00000281 00000282 00000283 00000284 00000285 00000286 00000287 /********************\ |* BUTTONS *| \********************/ private void btnNew_Click(object sender. EventArgs e) { new_network().IsBusy) { if (network != null) { if (!network.Trained) { bw.ShowDialog(). EventArgs e) { if (network != null) { if (network.RunWorkerAsync().Trained) Page 89 of 225 . } } } } private void btnTest_Click(object sender. if (!bw. workingdialog. EventArgs e) { //Shouldn't be busy if the user managed to click this //but just make 100% certain. } private void btnTrain_Click(object sender.

Text = network. txtOutput. } } } } } #endregion #region bg_worker /********************\ |* BG WORKER *| \********************/ private void bw_DoWork(object sender. if (network != null) { success = network. if (success) { txtOutput. DoWorkEventArgs e) { bool success = true.Test().Length.Clear(). txtOutput. txtOutput. txtOutput.SelectionStart = txtOutput.Tested) { bool success = network.Refresh().ScrollToCaret().ReportData.Train().Text. } Page 90 of 225 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000288 00000289 00000290 00000291 00000292 00000293 00000294 00000295 00000296 00000297 00000298 00000299 00000300 00000301 00000302 00000303 00000304 00000305 00000306 00000307 00000308 00000309 00000310 00000311 00000312 00000313 00000314 00000315 00000316 00000317 { if (!network.

Text. MessageBoxIcon.ToString(). MessageBoxButtons. RunWorkerCompletedEventArgs e) { bool success.ScrollToCaret().SelectionStart = txtOutput.OK.getGraphData()).Show("Error during network train: " + e.Error != null) { MessageBox.Error).Clear().Result.Error).Length. txtOutput. MessageBoxIcon. } Page 91 of 225 .ReportData. txtOutput.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000318 00000319 00000320 00000321 00000322 00000323 00000324 00000325 00000326 00000327 00000328 00000329 00000330 00000331 00000332 00000333 00000334 00000335 00000336 00000337 00000338 00000339 00000340 00000341 00000342 00000343 00000344 00000345 00000346 00000347 e. if (success) { txtOutput.Text = network. txtOutput.Result = success. plotgraphsMSE(network.OK.Show("Error training network!". "Error". } success = (bool) e. MessageBoxButtons. } else { MessageBox. //Check for errors if (e.Error. "Error training network!". } private void bw_RunWorkerCompleted(object sender.

EventArgs e) { //Run the new neural net wizard new_network(). System.1. System.“Wiz1. System.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000348 00000349 00000350 00000351 00000352 00000353 00000354 00000355 00000356 00000357 00000358 00000359 00000360 00000361 00000362 00000363 00000364 00000365 00000366 } //Remove the dialog locking out the main form //and showing that we're currently working workingdialog.Hide(). } #endregion } 9.Linq. } #endregion #region misc /********************\ |* MISC *| \********************/ private void FrmMain_Shown(object sender.cs” 00000001 00000002 00000003 00000004 00000005 00000006 using using using using using using System. System. System.3 .Drawing.1.Collections. Page 92 of 225 .Generic.Data.ComponentModel.

private int algorithm. using _05025397.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 00000030 00000031 00000032 00000033 00000034 00000035 00000036 using System. Page 93 of 225 . BP.Windows. using System. public event NetworkUpdateHandler NetworkUpdated.NetworkUpdateEventArgs e). CUSTOM }. private int Trainset. CC }.Forms.Controller { #region enums /********************\ |* ENUMS *| \********************/ enum Training { GA.Text. FISHER. enum Dataset { XOR. #endregion public partial class Wiz1 : Form { #region data /********************\ |* DATA *| \********************/ public delegate void NetworkUpdateHandler(object sender. Classes. namespace _05025397.

} #endregion #region control_events /********************\ |* CONTROL EVENTS *| \********************/ #region buttons /********************\ |* BUTTONS *| \********************/ private void btnCancel_Click(object sender.Close().XOR. algorithm = (int)Training.GA.icon. EventArgs e) Page 94 of 225 .Icon = Properties. this.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 00000060 00000061 00000062 00000063 00000064 00000065 00000066 #endregion data #region constructor /********************\ |* CONSTRUCTOR *| \********************/ public Wiz1() { InitializeComponent().Resources. Trainset = (int)Dataset. } private void rBtnBackprop_CheckedChanged(object sender. EventArgs e) { this.

Resources. panCC. lblBP.Visible = false.Visible = true.Visible = true. Page 95 of 225 . lblBP.IMAGEBackProp. panCC.Checked == true) { picAlg. lblCC.Visible = false.Properties.Visible = true.Visible = false.CC.Checked == true) { picAlg. panGeneticAlgorithm.Visible = false. algorithm = (int)Training. lblGA.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 00000090 00000091 00000092 00000093 00000094 00000095 00000096 { if (rBtnBackprop. panBackprop. panBackprop.Visible = false.BP. EventArgs e) { if (rBtnCascadeCorrelation. lblGA.Resources. lblCC. panGeneticAlgorithm. algorithm = (int)Training.Image = _05025397.IMAGECascadeCorrelation.Visible = false.Image = _05025397.Properties.Visible = false. } } private void rBtnCascadeCorrelation_CheckedChanged(object sender.Visible = false.Visible = true.

Checked == true) { picAlg.1). EventArgs e) { if (rBtnGeneticAlgorithm.Properties.GA. panBackprop. panCC. lblBP. } } private void btnBack_Click(object sender. lblGA.Image = _05025397. lblCC.IMAGEGeneticAlgorithm. panGeneticAlgorithm. EventArgs e) { Page 96 of 225 .Visible = true.Resources.SelectedIndex . algorithm = (int)Training.Visible = false. } } private void btnNext_Click(object sender. EventArgs e) { if (tabControl1.Visible = false.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 00000117 00000118 00000119 00000120 00000121 00000122 00000123 00000124 00000125 00000126 } } private void rBtnGeneticAlgorithm_CheckedChanged(object sender.SelectTab(tabControl1.SelectedIndex > 0) { tabControl1.Visible = false.Visible = false.Visible = true.

Checked.GA_Crossover. check. check.Text.Text. txtCC_TargetError. Trainset).Text.Text. chkGA_IgnoreTarget.Checked. Page 97 of 225 . txtGA_HiddenNodes.GA_GenLimit.TrainingAlgorithm. check. if (check. check. txtCC_Report.Text.Text. txtBP_HiddenNodes.GA_PopSize.Text. chkBP_IgnoreTarget.Text. txtGA_Terr. txtBP_Terr. } else { VerifyData check = new VerifyData (algorithm. txtGA_ProbX.GA_ITerr.GA_Terr.Text. txtBP_LearnRate. check.CC_ITerr.Verified) { Classes. check. txtCC_LearningRate.Text.CC_MaxNeurons. check.BP_EpochLimit. txtGA_PopSize.TabCount . txtCC_MaxNeurons.Text.NetworkUpdateEventArgs args = new Classes. txtBP_Report.GA_Mutation. check. txtGA_Report. check. txtGA_MaxGen.SelectedIndex < tabControl1.NetworkUpdateEventArgs (check. txtGA_ProbM.GA_HiddenL.CC_Terr. check. CC_ITerr.Text. check. check.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 00000146 00000147 00000148 00000149 00000150 00000151 00000152 00000153 00000154 00000155 00000156 if (tabControl1.BP_HiddenL.Text.Checked.Text.Text. check.1) { tabControl1. txtBP_EpochLimit.SelectTab(tabControl1.CC_Reports.GA_Reports.Text. check. check.SelectedIndex + 1).CC_LearningRate.

Trainset = (int)Dataset.BP_Reports. } else { MessageBox. MessageBoxIcon. args).Exclamation).Show("Error Checking Data: Data could not be verified". } } } #endregion #region checkboxes /********************\ |* CHECKBOXES *| \********************/ private void chkDatasetFisher_CheckedChanged(object sender.BP_Terr.Resources.Image = _05025397. check.BP_LearningRate.Dispose(). check.Properties.Checked) { picDataset.TrDataSet). } } private void chkDatasetXOR_CheckedChanged(object sender.BP_ITerr. NetworkUpdated(this.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 00000168 00000169 00000170 00000171 00000172 00000173 00000174 00000175 00000176 00000177 00000178 00000179 00000180 00000181 00000182 00000183 00000184 00000185 00000186 check. EventArgs e) { if (chkDatasetFisher. check.FISHER. "Error Checking Data".IMAGEIrisFlowers. MessageBoxButtons.OK. EventArgs e) Page 98 of 225 . check. this.

IMAGEQuestionMark.Checked) { picDataset. } } private void chkDatasetCustom_CheckedChanged(object sender. } } #endregion #region misc /********************\ |* MISC *| \********************/ private void tabControl1_SelectedIndexChanged(object sender.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000187 00000188 00000189 00000190 00000191 00000192 00000193 00000194 00000195 00000196 00000197 00000198 00000199 00000200 00000201 00000202 00000203 00000204 00000205 00000206 00000207 00000208 00000209 00000210 00000211 00000212 00000213 00000214 00000215 00000216 { if (chkDatasetXOR.Properties.IMAGEXOR.Properties.Focus(). EventArgs e) { if (tabControl1.XOR.Image = _05025397.Resources.Checked) { picDataset.Image = _05025397.Resources.SelectedIndex == 1) { if (panCC. } Page 99 of 225 . EventArgs e) { if (chkDatasetCustom.Visible == true) { txtCC_TargetError. Trainset = (int)Dataset.CUSTOM. Trainset = (int)Dataset.

Icon = Properties. EventArgs e) { this. tabControl1.icon.Visible == true) { txtBP_EpochLimit.Focus(). } if (panGeneticAlgorithm.SelectTab(0).Focus().William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000217 00000218 00000219 00000220 00000221 00000222 00000223 00000224 00000225 00000226 00000227 00000228 00000229 00000230 00000231 00000232 00000233 00000234 00000235 00000236 } if (panBackprop. } #endregion #endregion } Page 100 of 225 .Visible == true) { txtGA_PopSize. } } } private void Wiz1_Load(object sender.Resources.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9.Generic.Forms. System.cs” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 using using using using using using using using System.Controller { public partial class FrmAbout : Form { #region constructor /********************\ |* CONSTRUCTOR *| \********************/ public FrmAbout() { InitializeComponent(). EventArgs e) { Page 101 of 225 . System. System. System.Collections. namespace _05025397. System.1.Data.ComponentModel. System. } #endregion #region form /********************\ |* FORM *| \********************/ private void FrmAbout_Load(object sender.Text.Drawing.4 .Linq.Windows.1.“FrmAbout. System.

Icon = Properties.Text).LICENCE_LGPL21. "Undefined Application".Text = Application. MessageBoxIcon. LinkLabelLinkClickedEventArgs e) Page 102 of 225 .LinkVisited = true.OK. } catch { MessageBox.linkLabel1.Process.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 this. } #endregion #region linklabels /********************\ |* LINKLABELS *| \********************/ private void linkLabel1_LinkClicked(object sender. MessageBoxButtons.Resources.ProductVersion.Show("No e-mail program defined.icon.Information).Text = Properties. label5. System.Diagnostics. textBox5. } } private void linkLabel4_LinkClicked(object sender. LinkLabelLinkClickedEventArgs e) { // Specify that the link was visited.Resources. try { // Navigate to a URL. this.Start("mailto:" + linkLabel1.".

} catch { MessageBox. } catch { MessageBox.".Text).Diagnostics. LinkLabelLinkClickedEventArgs e) Page 103 of 225 .Diagnostics.Error). "Error". "Error".OK. LinkLabelLinkClickedEventArgs e) { try { System. MessageBoxIcon.Text).Start (linkLabel4.".Process.Process. MessageBoxButtons.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 { try { System.Start (linkLabel5. MessageBoxButtons. } } private void linkLabel6_LinkClicked(object sender.Show("Error opening link. MessageBoxIcon.Show("Error opening link.Error). } } private void linkLabel5_LinkClicked(object sender.OK.

Show("Error opening link.Error).Start ("http://www.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 00000117 00000118 00000119 { try { System.Diagnostics. "Error". LinkLabelLinkClickedEventArgs e) Page 104 of 225 .".Start (linkLabel6. } catch { MessageBox.Process.Diagnostics.OK.co.". } } private void linkLabel2_LinkClicked(object sender. MessageBoxIcon. MessageBoxButtons. MessageBoxButtons.uk").Process.snikks. LinkLabelLinkClickedEventArgs e) { try { System. "Error". MessageBoxIcon. } } private void linkLabel7_LinkClicked(object sender.OK.Error).Show("Error opening link. } catch { MessageBox.Text).

org/licenses/old-licenses/lgpl-2.Visible = false.Visible = true. } } #endregion #region buttons /********************\ |* BUTTONS *| \********************/ private void btnDatasets_Click(object sender. groupBox2. "Error".Process. } catch { MessageBox. } else Page 105 of 225 .Visible != true) { groupBox1. MessageBoxIcon.Start ("http://www.Error). pictureBox1.OK.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 00000146 00000147 00000148 00000149 { try { System.Diagnostics.html").Visible = false.1.Show("Error opening link. tabControl1.". EventArgs e) { if (tabControl1.Visible = false.gnu. MessageBoxButtons.

Visible } } private void btnLicences_Click(object sender.Visible != true) { groupBox2. = true. EventArgs e) { if (groupBox2.Dispose().Visible = groupBox2.Visible = true. Page 106 of 225 .Visible = tabControl1.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 00000168 00000169 00000170 00000171 00000172 00000173 00000174 00000175 00000176 00000177 00000178 00000179 { groupBox1.Visible = false. groupBox2. pictureBox1. } else { groupBox1.Visible = true. false.Visible pictureBox1. pictureBox1. = false. tabControl1.Visible = true. groupBox1. tabControl1. EventArgs e) { this.Visible = false.Visible = false. } } private void btnOkay_Click(object sender.Visible = false.Visible = false. } true.

LICENCE_LGPL21.Resources.LICENCE_GPL30.Checked == true) textBox5.Text = Properties. EventArgs e) { if (radioButton2. } #endregion } Page 107 of 225 . EventArgs e) { if (radioButton1.Resources.Checked == true) textBox5.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000180 00000181 00000182 00000183 00000184 00000185 00000186 00000187 00000188 00000189 00000190 00000191 00000192 00000193 00000194 00000195 00000196 00000197 00000198 00000199 00000200 00000201 00000202 00000203 00000204 00000205 } #endregion #region radiobuttons /********************\ |* RADIOBUTTONS *| \********************/ private void radioButton1_CheckedChanged(object sender.Text = Properties. } private void radioButton3_CheckedChanged(object sender.Text = Properties. } private void radioButton2_CheckedChanged(object sender.Checked == true) textBox5.LICENCE_LGPL30.Resources. EventArgs e) { if (radioButton3.

Linq.5 .ComponentModel. System. namespace _05025397.“Working.Beep.Generic.cs” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 using using using using using using using using System.Controller { public partial class Working : Form { #region functions /********************\ |* FUNCTIONS *| \********************/ public Working() { InitializeComponent().Icon = Properties.Resources. System. System.SystemSounds.1. System.Text.Forms.Drawing. EventArgs e) { System. System. } private void Working_Click(object sender.Windows. System.Media.Collections. } #endregion } Page 108 of 225 . this.icon. System.Data.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9.Play().1.

Forms. System.2 .Generic. System.Linq.Windows.Controller 9.1 .Text.cs” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 using using using using using System. //Dataset identifer //0 = XOR Page 109 of 225 .Collections. System. //Training Algorithm Identifier //1 = genetic algorith //2 = backprop //3 = cascade correlation private int trainingalg.Controller { public class VerifyData { #region data /********************\ |* DATA *| \********************/ //Dataverified variable private bool verified.1.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000030 } 9.2. System.1. namespace _05025397.“VerifyData.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000026 00000027 00000028 00000029 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 //1 = FISHERS IRIS DATA //2 = CUSTOM private int _TDataSet. private int _GA_HiddenL. private double _GA_Terr. private int _GA_Mutation. private int _GA_Crossover. private int _GA_Reports. private int _CC_Reports. private int _CC_MaxNeurons. //BackProp private int _BP_EpochLimit. //Cascade Correlation private double _CC_Terr. private bool _GA_ITerr. private double _CC_LearningRate. private int _GA_GenLimit. private int _BP_HiddenL. private bool _CC_ITerr. //Genetic Algorithm private int _GA_PopSize. private bool _BP_ITerr. #endregion Page 110 of 225 . private double _BP_LearningRate. private double _BP_Terr. private int _BP_Reports.

//DataSet _TDataSet = TData_Set. string GAGenLimit. //Algorithm trainingalg = Algorithm. string BPReports.XOR && _TDataSet != Page 111 of 225 . will be changed to //false if any data verification errors occur verified = true. bool CCITerr. string GACrossOver. string BPTerr. string CCTerr. string GAMutation. bool GAITerr. string CCReports. if (_TDataSet != (int)Dataset. string BPEpochLimit. string CCLearningRate. bool BPITerr. int TData_Set) { //Initially set to true. string BPHiddenL. string GAHiddenL.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000056 00000057 00000058 00000059 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 #region constructor /********************\ |* CONSTRUCTOR *| \********************/ public VerifyData (int Algorithm. string GAReports. string BPLearningRate. string GAPopSize. string GATerr. string CCMaxNeurons.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000086 00000087 00000088 00000089 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 (int)Dataset. } catch (FormatException) { MessageBox.FISHER && _TDataSet != (int)Dataset.Error). MessageBoxIcon.Show ("Parsing Error: Please check supplied cascade" + "correlation data values". if (verified) { if (trainingalg == 2) { try { //Cascade correlation _CC_Terr = Convert. _CC_LearningRate = Convert. _CC_Reports = Convert. verified = false.ToInt32(CCReports).ToSingle(CCLearningRate).OK.CUSTOM) verified = false. MessageBoxButtons. "Parsing Error".ToInt32(CCMaxNeurons). _CC_MaxNeurons = Convert.ToSingle(CCTerr). _CC_ITerr = CCITerr. } } } Page 112 of 225 .

Show ("Parsing Error: Please check supplied" +" genetic algorithm data values". } catch (FormatException) { MessageBox.OK.ToInt32(GAMutation). _GA_HiddenL = Convert. _GA_GenLimit = Convert. _GA_ITerr = GAITerr. "Parsing Error".ToInt32(GAGenLimit).ToInt32(GAReports).ToSingle(GATerr).ToInt32(GACrossOver).ToInt32(GAPopSize).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000116 00000117 00000118 00000119 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 if (verified) { if (trainingalg == 0) { try { //Genetic algorithm _GA_PopSize = Convert. verified = false. MessageBoxIcon. MessageBoxButtons. _GA_Reports = Convert. _GA_Mutation = Convert.Error). } } } if (verified) Page 113 of 225 . _GA_Terr = Convert.ToInt32(GAHiddenL). _GA_Crossover = Convert.

ToInt32(BPReports).ToInt32(BPHiddenL).ToDouble(BPTerr). } } } if (verified) { verified = checkforsanity(). _BP_LearningRate = Convert. _BP_HiddenL = Convert. _BP_Reports = Convert. _BP_Terr = Convert.Show ("Parsing Error: Please check " + "supplied back propagation data values". } catch (FormatException) { MessageBox. verified = false. _BP_ITerr = BPITerr.OK.ToInt32(BPEpochLimit).ToDouble(BPLearningRate).Error). } Page 114 of 225 . MessageBoxButtons. "Parsing Error".William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000146 00000147 00000148 00000149 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 00000168 00000169 00000170 00000171 00000172 00000173 00000174 00000175 { if (trainingalg == 1) { try { //Back-propagation _BP_EpochLimit = Convert. MessageBoxIcon.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000176 00000177 00000178 00000179 00000180 00000181 00000182 00000183 00000184 00000185 00000186 00000187 00000188 00000189 00000190 00000191 00000192 00000193 00000194 00000195 00000196 00000197 00000198 00000199 00000200 00000201 00000202 00000203 00000204 00000205 } #endregion #region internal /********************\ |* INTERNAL *| \********************/ private bool checkGAsanity() { bool sanity = true. if (_GA_GenLimit < 1) sanity = false. } if (_GA_Terr < 0 || _GA_Terr > 1) sanity = false. if (_GA_Crossover < 0 || _GA_Crossover > 100) { sanity = false. Page 115 of 225 . } if (_GA_Mutation < 0 || _GA_Mutation > 100) { sanity = false. if (_GA_PopSize < 3) sanity = false.

if (_GA_HiddenL < 1) sanity = false. if (!sanity) MessageBox. if (_BP_EpochLimit < 1) sanity = false. "Sanity Check Fail". MessageBoxButtons. return sanity. if (_BP_Reports < 1 || _BP_Reports > _BP_EpochLimit) sanity = false.Show ("Sanity Check Fail: Please check supplied " + "genetic algorithm data values". MessageBoxIcon. if (_BP_HiddenL < 1) sanity = false.Error). } private bool checkBPsanity() { bool sanity = true.OK. Page 116 of 225 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000206 00000207 00000208 00000209 00000210 00000211 00000212 00000213 00000214 00000215 00000216 00000217 00000218 00000219 00000220 00000221 00000222 00000223 00000224 00000225 00000226 00000227 00000228 00000229 00000230 00000231 00000232 00000233 00000234 00000235 if (_GA_Reports < 1 || _GA_Reports > _GA_GenLimit) sanity = false.

if (!sanity) MessageBox. "Sanity Check Fail".Show ("Sanity Check Fail: Please check supplied " + "back propagation data values".William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000236 00000237 00000238 00000239 00000240 00000241 00000242 00000243 00000244 00000245 00000246 00000247 00000248 00000249 00000250 00000251 00000252 00000253 00000254 00000255 00000256 00000257 00000258 00000259 00000260 00000261 00000262 00000263 00000264 00000265 if (_BP_Terr < 0 || _BP_Terr > 1) sanity = false. } private bool checkCCsanity() { bool sanity = true. MessageBoxButtons. return sanity. if (_BP_LearningRate < 0 || _BP_LearningRate > 1) sanity = false. if (_CC_Reports < 1 || _CC_Reports > _CC_MaxNeurons) sanity = false. Page 117 of 225 . MessageBoxIcon. if (_CC_Terr < 0 || _CC_Terr > 1) sanity = false. if (_CC_MaxNeurons < 1) sanity = false.Error).OK.

OK. } private bool checkforsanity() { //Some of the values that are allowed through here would not produce //good results . MessageBoxIcon.however I don't want to prevent any flexibility. MessageBoxButtons.Error). sanity = true. only //halt any values from proceeding that are so wrong they could cause //serious errors later down the line. if (!sanity) MessageBox. if they're greater than 1 but less than 100 scale them appropriately. We're also doing some very minor processing of the probability values. return sanity. "Sanity Check Fail".Show ("Sanity Check Fail: Please check supplied " + "cascade correlation data values". // // // // // bool Basically we're making sure the data is clean before it goes anywhere near a DLL. if (trainingalg < 0 || trainingalg > 2) Page 118 of 225 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000266 00000267 00000268 00000269 00000270 00000271 00000272 00000273 00000274 00000275 00000276 00000277 00000278 00000279 00000280 00000281 00000282 00000283 00000284 00000285 00000286 00000287 00000288 00000289 00000290 00000291 00000292 00000293 00000294 00000295 if (_CC_LearningRate < 0 || _CC_LearningRate > 1) sanity = false.

MessageBoxIcon. } Page 119 of 225 . MessageBox.OK. MessageBoxButtons. } if (trainingalg == 2) { sanity = checkCCsanity().Error). "Sanity Check Fail".William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000296 00000297 00000298 00000299 00000300 00000301 00000302 00000303 00000304 00000305 00000306 00000307 00000308 00000309 00000310 00000311 00000312 00000313 00000314 00000315 00000316 00000317 00000318 00000319 00000320 00000321 00000322 00000323 00000324 00000325 { sanity = false. } if (sanity) { if (trainingalg == 0) { sanity = checkGAsanity(). } if (trainingalg == 1) { sanity = checkBPsanity(). } } return sanity.Show ("Sanity Check Fail: Training Algorithm not " + "selected or selected incorrectly".

} } public double CC_LearningRate { get { return _CC_LearningRate. } } public bool CC_ITerr { get { return _CC_ITerr.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000326 00000327 00000328 00000329 00000330 00000331 00000332 00000333 00000334 00000335 00000336 00000337 00000338 00000339 00000340 00000341 00000342 00000343 00000344 00000345 00000346 00000347 00000348 00000349 00000350 00000351 00000352 00000353 00000354 00000355 #endregion #region getters/setters /********************\ |* EXTERNAL *| \********************/ #region cascade_correlation //Cascade Correlation public double CC_Terr { get { return _CC_Terr. } } #endregion #region genetic_algorithm //Genetic Algorithm public int GA_PopSize { get { return _GA_PopSize. } } public int GA_Crossover { get { return _GA_Crossover. } } public int CC_MaxNeurons { get { return _CC_MaxNeurons. } } public int CC_Reports { get { return _CC_Reports. } } public int GA_GenLimit { get { return _GA_GenLimit. } } public int GA_Mutation Page 120 of 225 .

} } _GA_HiddenL. } } public bool BP_ITerr { get { return _BP_ITerr. } } _GA_Reports. } } public double BP_Terr { get { return _BP_Terr. } } #endregion #endregion #region external //Verification Page 121 of 225 . } } public int BP_HiddenL { get { return _BP_HiddenL. } } public int BP_Reports { get { return _BP_Reports.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000356 00000357 00000358 00000359 00000360 00000361 00000362 00000363 00000364 00000365 00000366 00000367 00000368 00000369 00000370 00000371 00000372 00000373 00000374 00000375 00000376 00000377 00000378 00000379 00000380 00000381 00000382 00000383 00000384 00000385 { get { return public double GA_Terr { get { return public int GA_Reports { get { return public int GA_HiddenL { get { return public bool GA_ITerr { get { return #endregion _GA_Mutation. } } _GA_ITerr. } } public double BP_LearningRate { get { return _BP_LearningRate. } } _GA_Terr. } } #region back_propagation //Back propagation public int BP_EpochLimit { get { return _BP_EpochLimit.

} } //TrainingAlgorithm public int TrainingAlgorithm { get { return trainingalg.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000386 00000387 00000388 00000389 00000390 00000391 00000392 00000393 00000394 00000395 00000396 00000397 00000398 } public bool Verified { get { return verified. } } #endregion } Page 122 of 225 . } } //Dataset public int TrDataSet { get { return _TDataSet.

cs” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 using using using using using using System.Collections.Windows. private ArrayList aldata.Collections. private ArrayList trinputs. private string[] data.Forms. System. private ArrayList troutputs. System. System.Text. private ArrayList teoutputs. namespace _05025397.Controller { public class DatasetParser { #region data /********************\ |* DATA *| \********************/ private bool _Verified. private ArrayList teinputs. private string[] test. private ArrayList altest.2. Page 123 of 225 . System.Linq. System.“DatasetParser.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9.1.Generic.2 .

txtTest. RichTextBox txtTest) { //Data init bool Verified = true. '.'.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 //private Scaler scale.' }.Text = txtTest. //Strip carriage returns and trim txtData. aldata = new ArrayList(). private double Ninputs.Text.Replace("\r". ""). trinputs = new ArrayList().Text = txtTest. troutputs = new ArrayList().Text.Text = txtData. '\r'.Text. #endregion #region constructor /********************\ |* CONSTRUCTOR *| \********************/ public DatasetParser(RichTextBox txtData. private double Noutputs. ' '. '\t'. "").Replace("\r". Page 124 of 225 . char[] delimiterChars = { '\n'. altest = new ArrayList().Trim().Trim(). '. teoutputs = new ArrayList().Text. private double Niterations. txtData.Text = txtData. teinputs = new ArrayList(). txtTest.

Verified = parseTestingData().William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 //Split at the chars specified above data = txtData. i++) Page 125 of 225 .Convert.Length. Ninputs = System.Text.ToDouble(data[1]).ToDouble(data[2]).Text. i < data. _Verified = Verified. //Get the number of iterations. number of inputs //and number of outputs Niterations = System. } #endregion #region internal_functions /********************\ |* INTERNAL *| \********************/ private bool parseTrainingData() { bool success = true.Split(delimiterChars). //parse the data into appropriately //structured arrays Verified = parseTrainingData(). try { for (int i = 3.ToDouble(data[0]). test = txtTest.Convert. Noutputs = System.Convert.Split(delimiterChars).

RemoveAt(0).Show("Error parsing training data". aldata.Add(System. } private bool parseTestingData() { bool success = true. } for (int j = 0. aldata. "Error". j++) { troutputs.Count > 0) { for (int i = 0. j < Noutputs.OK.Add(aldata[0]). MessageBoxIcon.RemoveAt(0). Page 126 of 225 . MessageBoxButtons. i++) { trinputs.Convert. } } } catch { MessageBox.Add(aldata[0]).Error). } while (aldata.ToDouble(data[i])).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 00000117 00000118 00000119 { aldata. i < Ninputs. } return success. success = false.

i++) { altest.Add(System.Show("Error parsing testing data". j < Noutputs.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 00000146 00000147 00000148 00000149 try { for (int i = 3.OK.Error). } } } catch { MessageBox.Add(altest[0]).RemoveAt(0).ToDouble(data[i])). } return success. altest. } while (altest. j++) { teoutputs. "Error".Convert. success = false. i++) { teinputs. i < data. MessageBoxIcon.Length.Count > 0) { for (int i = 0. altest. } for (int j = 0. i < Ninputs. MessageBoxButtons.Add(altest[0]). } Page 127 of 225 .RemoveAt(0).

Count / (int) Ninputs][]. j < trinputs. k < Ninputs.Count / (int) Noutputs][]. troutputdata = new double[troutputs. i++) trinputdata[i] = new double[(int) Ninputs]. for (int j = 0. out double[][] troutputdata) { trinputdata = new double[trinputs. j+=(int) Ninputs. for (int i = 0. l = 0. out double Ninp.Count. } public void GetTrainingData (out double[][] trinputdata. k++) { Page 128 of 225 . for (int i = 0. l++) { for (int k = 0. iter = Niterations. out double Nout) { Ninp = Ninputs. i < (trinputs.Count / (int)Ninputs).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 00000168 00000169 00000170 00000171 00000172 00000173 00000174 00000175 00000176 00000177 00000178 00000179 #endregion #region external_functions /********************\ |* EXTERNAL *| \********************/ public void GetStructure(out double iter. i++) troutputdata[i] = new double[(int) Noutputs]. i < (troutputs. Nout = Noutputs.Count / (int)Noutputs).

for (int j = 0. for (int i = 0. k < Ninputs. l++) { for (int k = 0. } } } public void GetTestingData (out double[][] teinputdata. i++) teinputdata[i] = new double[(int)Ninputs]. j < troutputs. teoutputdata = new double[teoutputs.Count / (int)Noutputs). l++) { for (int k = 0. k < Noutputs.Count. l = 0. out double[][] teoutputdata) { teinputdata = new double[teinputs. j += (int)Noutputs. j += (int)Ninputs. k].William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000180 00000181 00000182 00000183 00000184 00000185 00000186 00000187 00000188 00000189 00000190 00000191 00000192 00000193 00000194 00000195 00000196 00000197 00000198 00000199 00000200 00000201 00000202 00000203 00000204 00000205 00000206 00000207 00000208 00000209 trinputdata[l][k] = (double) trinputs[j + } } for (int j = 0. Page 129 of 225 . for (int i = 0. i++) teoutputdata[i] = new double[(int)Noutputs].Count / (int)Ninputs][]. l = 0.Count.Count / (int)Ninputs). i < (teinputs.Count / (int)Noutputs][]. j < teinputs. i < (teoutputs. k++) { teinputdata[l][k] = (double)teinputs[j + k]. k++) { troutputdata[l][k] = (double)troutputs[j + k].

l++) { for (int k = 0. j += (int)Noutputs.1. teoutputdata = teoutputs. } } #endregion } 9.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000210 00000211 00000212 00000213 00000214 00000215 00000216 00000217 00000218 00000219 00000220 00000221 00000222 00000223 00000224 00000225 00000226 00000227 00000228 00000229 00000230 00000231 00000232 00000233 00000234 00000235 00000236 } } } for (int j = 0.2. k++) { teoutputdata[l][k] = (double)teoutputs[j + k]. k < Noutputs. } } } public void GetTestingDataAL (out ArrayList teinputdata.3 . j < teoutputs. l = 0.“Algorithms\TrainingAlgorithm. } public bool Verified { get { return _Verified. Page 130 of 225 .Count.cs” 00000001 using System. out ArrayList teoutputdata) { teinputdata = teinputs.

Algorithms { public abstract class TrainingAlgorithm { #region data /********************\ |* DATA *| \********************/ private double _TargetError. System.Forms. protected bool _Tested. namespace _05025397.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 00000030 00000031 using using using using using using System.Text. private int _IterMax.Generic.Controller. ZedGraph.Windows. System. private string _ReportData. protected bool _ITerr. protected PointPairList BestErrList.Collections. protected int CurIt. protected bool _Trained.Linq. Page 131 of 225 . private int _ReportInterval. System. _05025397.

if (_ITerr) _TargetError = -1.0 //this is impossible to reach. bool ITerr) { _TargetError = Target_Error. so the //network will loop to completion. _ReportInterval = Reports. //If the ignore target error is true //then set the target error to -1. _ITerr = ITerr. RichTextBox txtTest. } #endregion Page 132 of 225 . _IterMax = Iter_Max. int Iter_Max. BestErrList = new PointPairList(). RichTextBox txtData. int Reports.00. txtTest). #endregion #region constructor /********************\ |* CONSTRUCTOR *| \********************/ public TrainingAlgorithm(double Target_Error.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 00000060 00000061 protected Random Rand = new Random(). protected DatasetParser getdata. getdata = new DatasetParser(txtData.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 00000090 00000091 #region internal /********************\ |* INTERNAL *| \********************/ protected void Report(string output) { _ReportData += output. } get { return _TargetError. public PointPairList getGraphData() { return BestErrList. } } Page 133 of 225 . } public double TargetError { set { _TargetError = value. public abstract bool Test(). public abstract string network_details(). } #endregion #region external /********************\ |* EXTERNAL *| \********************/ public abstract bool Train().

} } public bool Trained { get { return _Trained. } } public int IterMax { get { return _IterMax. } } public bool Tested { get { return _Tested.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 00000117 00000118 00000119 } public int ReportInterval { get { return _ReportInterval. } } #endregion } Page 134 of 225 . } set { _Trained = value. } } public string ReportData { get { return _ReportData.

Windows.Linq.4 . FANN_Cascade CCNet.Algorithms { public class CascadeCorrelation : TrainingAlgorithm { #region data /********************\ |* DATA *| \********************/ private double _InputL. System. System.IO.Text. System. private double _OutputL.Collections.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9. FANN_Wrapper. _05025397. System. private double _LearnRate.Collections.“Algorithms\CascadeCorrelation.2.Generic.Forms. ZedGraph. #endregion #region constructor Page 135 of 225 . namespace _05025397.cs” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 using using using using using using using using using using System. System.Controller.1. System.

base. getdata. RichTextBox te) { Page 136 of 225 . txtData. } #endregion #region internal_functions /********************\ |* INTERNAL *| \********************/ private void saveDllData(RichTextBox tr. txtTest. RichTextBox txtData. int Iter_Max.ReportInterval. Iter_Max. saveDllData(txtData. ITerr) { _LearnRate = LearnRate. RichTextBox txtTest) : base(TargetError. base. txtTest). bool ITerr. base. double iterations.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 /********************\ |* CONSTRUCTOR *| \********************/ public CascadeCorrelation(double LearnRate. int Reports.IterMax). CCNet = new FANN_Cascade((int)_InputL.TargetError. Reports. out _InputL. _LearnRate. double TargetError. (int)_OutputL.GetStructure(out iterations. out _OutputL).

tw.Append( string.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 TextWriter tw = new StreamWriter("dlltrdata.Append( Page 137 of 225 .Append( string. _LearnRate)). } #endregion #region external_functions /********************\ |* EXTERNAL *| \********************/ public override string network_details() { StringBuilder output = new StringBuilder().Text).dat"). output. tw. (int)_InputL)). output.00000}".Write(tr.dat").Format ("\nLearning Rate: {0:0. output. output.Text).Write(te. tw. tw = new StreamWriter("dlltedata.Format ("\t\t\t\t\tOutput Layer: {0:d}".Format ("Network Details: \n Input Layer: {0:d}".Close().Append( string. tw. (int)_OutputL)).Close().

} public override bool Train() { bool success = true.ReportInterval)). output. return output._tr_output).Format ("\t\t\tTarget Error: {0:0.Format ("\nReport every {0:d} generations.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 00000117 00000118 00000119 string.Format ("\t\t\t\t\t\tIgnore target error " + "(process all iterations): {0}".TargetError)).00000}".Append( string. output. output.Append( string. } catch { Page 138 of 225 . _ITerr)).Format ("\t\tMaximum Nodes: {0:d}".Append( string. Report(CCNet. try { success = CCNet.ToString(). base. base.Train(). base.".IterMax)).

OK. try { double[] besterrlist = CCNet.besterrorlist.Show("Error getting graph data from DLL".Error). base. for (int i = 0.Show("Error running Cascade training!". i++) { base.Add((double)i. MessageBoxIcon. return success. } public bool getGraphDataFromDll() { bool success = true.OK. "Error".William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 00000146 00000147 00000148 00000149 MessageBox. MessageBoxIcon.BestErrList. MessageBoxButtons. i < besterrlist. } Page 139 of 225 . besterrlist[i]). success = false.Error).Length. MessageBoxButtons. success = false._Trained = success. } } catch { MessageBox. } success = getGraphDataFromDll(). "Error".

} base. MessageBoxButtons.OK. try { success = CCNet. return success. } #endregion } Page 140 of 225 . "Error"._Tested = success. Report(CCNet.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 00000168 00000169 00000170 00000171 00000172 00000173 } return success.Error). } public override bool Test() { bool success = true.Test(). } catch { MessageBox. MessageBoxIcon._te_output). success = false.Show("Error running Cascade testing!".

cs” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 using using using using using using using using System.Collections. System.Collections. private int _Crossover. _05025397.Text.2.Controller. private int _Mutation.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9. System. namespace _05025397.5 .Forms. //Network Data private double _InputL.1.“Algorithms\GeneticAlgorithm. System.Windows. System. private int _HiddenL.Algorithms { public class GeneticAlgorithm : TrainingAlgorithm { #region data /********************\ |* DATA *| \********************/ //GA Data private int _PopSize.Linq.Generic. private double _OutputL. //Population Page 141 of 225 . ZedGraph. System.

txtData. double TargetError. getdata. int Mutation. Iter_Max. out _OutputL). #endregion #region constructor /********************\ |* CONSTRUCTOR *| \********************/ public GeneticAlgorithm(int PopSize. int Reports. _Mutation = Mutation. bool ITerr. int Crossover. RichTextBox txtData. Reports. ITerr) { _PopSize = PopSize. RichTextBox txtTest) : base(TargetError. //Init Network params double iterations. out _InputL. i < _PopSize. int HiddenL. _Crossover = Crossover.Add Page 142 of 225 . int Iter_Max.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 ArrayList Population. i++) { Population. _HiddenL = HiddenL. txtTest. //Create new population references Population = new ArrayList(_PopSize). for (int i = 0.GetStructure(out iterations.

out results). double[][] results. getdata. _HiddenL.GANetwork) Population[j]).getMsQE(inputs. double[][] inputs. Rand. } //Sort the functions according to fitness now. j++) { ((Model.Next())). results).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 (new Model. i < base. success = sort_fitnesses().GANetwork( (int)_InputL.IterMax.GetScaledTrainingData(out inputs. } } #endregion #region internal_functions /********************\ |* INTERNAL *| \********************/ private bool run() { bool success = true. (int)_OutputL. Page 143 of 225 . i++) { //Get the msqe for (int j = 0. j < _PopSize. for (int i = 0.

BestErrList.").MSqE + ".William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 00000117 00000118 00000119 //Report if we're at a report iteration //Also update the besterrlist (for graphing) if (base.CurIt++. base.MSqE).\n").GANetwork)Population[0]). } success = build_generation(). break.Report("Best error at iteration [ " + i + " ] was " + ((Model.GANetwork)Population[0]). } private bool sort_fitnesses() { Page 144 of 225 . } return success. base.CurIt == ReportInterval) { base. ((Model.CurIt = 0.Add ((double)i. } base.MSqE <= base. //Check for having reached the target if (((Model.TargetError) { Report("\nNetwork matching or improving upon target error"+ "found at iteration [ " + i + " ].GANetwork)Population[0]).

double[] weightarray2. //It selects two points from the start of the population //Then applies crossover if necessary (based on probability) Page 145 of 225 . i--) { //This algorithm is altered from previous versions. } return success. int selection1. selection2. } private bool build_generation() { bool success = true.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 00000146 00000147 00000148 00000149 bool success = true. } catch { MessageBox.Sort(). i > (int)(_PopSize / 2). try { Population. "Error". double[] weightarray1. success = false. for (int i = (_PopSize . MessageBoxButtons.1).OK.Show("Error sorting population". double[] weightarray_f. MessageBoxIcon.Error).

selection1 = Rand. //Just a quick error check if (weightarray1. weightarray1 = ((Model.getWeights().GANetwork)Population[selection1]).OK. MessageBoxButtons.". //Then it replaces the last position in the population //(ie. selection2 = Rand. the least fit) with this new string.Length].Next((int)_PopSize / 2).Error). all the way to the middle of the population. if (Rand. //Next iteration it replaces the next to last.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 00000168 00000169 00000170 00000171 00000172 00000173 00000174 00000175 00000176 00000177 00000178 00000179 //applies mutation if necessary (same again) //then if neither mutation or Xover have been applied //chooses position 1 or 2.GANetwork)Population[selection1]).Show("Error: 2D weight array length unequal. MessageBoxIcon.Next((int)_PopSize / 2).Length) { MessageBox. "Error". weightarray2 = ((Model. } try { weightarray_f = new double[weightarray1.Length != weightarray2.getWeights(). success = false. and //so on.Next(100) < _Crossover) { //Choose a random position in the weight array //to be our crossover point Page 146 of 225 .

5. (weightarray1.Next(100) < _Mutation) { //Jiggle the weights in the array //up or down by 0. j++) { if (j < XPos) { weightarray_f[j] = weightarray1[j].Length .0.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000180 00000181 00000182 00000183 00000184 00000185 00000186 00000187 00000188 00000189 00000190 00000191 00000192 00000193 00000194 00000195 00000196 00000197 00000198 00000199 00000200 00000201 00000202 00000203 00000204 00000205 00000206 00000207 00000208 00000209 int XPos = Rand. } } } if (Rand.Next(1. j < weightarray_f.setWeights(weightarray_f).5 for (int j = 0.Length. //Simple single point crossover of the weights for (int j = 0.NextDouble() . } else { weightarray_f[j] = weightarray2[j].Length.GANetwork)Population[i]). } Page 147 of 225 . } } //Set the weights of the current member of the //population we're on ((Model. j++) { weightarray_f[j] += Rand.1)). j < weightarray1.

Format("\t\t\tMutation Probability: {0:d}". output. } } return success. _Crossover)). output. _PopSize)).Format("Network Details: \n Population: {0:d}". output.Append( string.Error). MessageBoxIcon. success = false.Append( Page 148 of 225 . output.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000210 00000211 00000212 00000213 00000214 00000215 00000216 00000217 00000218 00000219 00000220 00000221 00000222 00000223 00000224 00000225 00000226 00000227 00000228 00000229 00000230 00000231 00000232 00000233 00000234 00000235 00000236 00000237 00000238 00000239 catch { MessageBox.Append( string.OK. } #endregion #region external_functions /********************\ |* EXTERNAL *| \********************/ public override string network_details() { //Add stringbuilder for efficiency here StringBuilder output = new StringBuilder().Format("\t\t\tCrossover Probability: {0:d}".Append( string. "Error". MessageBoxButtons.Show("Error building new generation". _Mutation)).

output. output. return output.Format("\t\t\tTarget Error: {0:0. base. base.00000}".Format("\nHidden layer: {0:d}". output.Format("\nReport every {0:d} generations. return success. success = run().TargetError)).Format("\t\t\t\t\t\t\tIgnore target error (process all iterations): {0}".Append( string.Append( string. } public override bool Train() { bool success = true.Append( string. _ITerr)).Append( string.ReportInterval)). _HiddenL)).IterMax)).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000240 00000241 00000242 00000243 00000244 00000245 00000246 00000247 00000248 00000249 00000250 00000251 00000252 00000253 00000254 00000255 00000256 00000257 00000258 00000259 00000260 00000261 00000262 00000263 00000264 00000265 00000266 00000267 00000268 00000269 string._Trained = success. output.". base.ToString(). } public override bool Test() Page 149 of 225 .Format("\t\t\tMaximum Generation: {0:d}". base.

try { for (int i = 0. i < tstinput. output. output. for (int j = 0. out tstoutput). getdata. for (int j = 0. netoutput[j])).GetOutput(). tstinput[i][j])). Page 150 of 225 . double[] netoutput.Append("\nOutput achieved was:").Format(" [{0:g}] ". netoutput = ((Model.Append("\n\n\t\t\t~~Network Testing~~"). StringBuilder output = new StringBuilder().Length.Length. j++) { output. i++) { ((Model.GetScaledTestingData(out tstinput.Length. } output.GANetwork)Population[0]).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000270 00000271 00000272 00000273 00000274 00000275 00000276 00000277 00000278 00000279 00000280 00000281 00000282 00000283 00000284 00000285 00000286 00000287 00000288 00000289 00000290 00000291 00000292 00000293 00000294 00000295 00000296 00000297 00000298 00000299 { bool success = true.Format(" [{0:g}] ".Append(string.Append("\nWith inputs:"). j < tstinput[i].GANetwork)Population[0]). double[][] tstinput.Append(string. double[][] tstoutput.Run(tstinput[i]). j++) { output. j < netoutput.

Append("\nMean Squared Error with these inputs and outputs is:"). (double) ((Model.Append("\nOutput Desired was:").MSqE)). "Error".Format(" [{0:g}] ".ToString()). MessageBoxIcon.getMsQE(tstinput. base.Error). MessageBoxButtons. output. } base.tstoutput). tstoutput[i][j]))._Tested = success. j < tstoutput[i].Report(output.OK.GANetwork)Population[0]).Length. j++) { output. for (int j = 0.Append( string. success = false. } catch { MessageBox. } } output.Show("Error Running Test.Append(string.".William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000300 00000301 00000302 00000303 00000304 00000305 00000306 00000307 00000308 00000309 00000310 00000311 00000312 00000313 00000314 00000315 00000316 00000317 00000318 00000319 00000320 00000321 00000322 00000323 00000324 00000325 00000326 00000327 00000328 } } output. } #endregion } Page 151 of 225 . ((Model.GANetwork)Population[0]). return success.Format ("{0:g}".

Forms.Linq. ZedGraph.Algorithms { public class BackProp : TrainingAlgorithm { #region data /********************\ |* DATA *| \********************/ private double _InputL.“Algorithms\BackProp.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9. FANN_BackProp BPnet. _05025397. private double _OutputL. System. System. private double _LearnRate.1. System. System.6 .Collections. namespace _05025397.Text.cs” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 using using using using using using using using using using System.IO. System.Generic.Windows. FANN_Wrapper. #endregion #region constructor Page 152 of 225 . System.Collections.2.Controller. private int _HiddenL.

ReportInterval. saveDllData(txtData. _LearnRate. base. HiddenL. double LearnRate. txtData. ITerr) { _HiddenL = HiddenL. txtTest). base.TargetError.IterMax). RichTextBox txtData. } #endregion #region internal_functions /********************\ Page 153 of 225 . txtTest. double TargetError. BPnet = new FANN_BackProp((int)_InputL. base. double iterations.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 /********************\ |* CONSTRUCTOR *| \********************/ public BackProp (int HiddenL. out _InputL. //Copy the training and testing data to the dat files //ready for the DLL's to access. bool ITerr. (int)_OutputL. out _OutputL). int Iter_Max. _LearnRate = LearnRate. int Reports. Iter_Max. getdata.GetStructure(out iterations. Reports. RichTextBox txtTest) : base(TargetError.

Text).Text).Append( string. RichTextBox te) { TextWriter tw = new StreamWriter("dlltrdata.Close(). output. tw = new StreamWriter("dlltedata. tw. tw. (int) _OutputL)). tw.Write(te. tw.Append( string.Format ("Network Details: \n Input Layer: {0:d}". output.Append( Page 154 of 225 . (int) _InputL)). } #endregion #region external_functions /********************\ |* EXTERNAL *| \********************/ public override string network_details() { StringBuilder output = new StringBuilder().William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 |* INTERNAL *| \********************/ private void saveDllData(RichTextBox tr.dat").dat").Close().Format ("\t\t\tOutput Layer: {0:d}".Write(tr. output.

Format ("\nLearning Rate: {0:0.00000}". return output. _LearnRate)). output.Append( string.". base.Append( string.Append( string.ToString().Format ("\t\t\t\t\t\tIgnore target error " + "(process all iterations): {0}".Append( string. output.00000}". output. output. output. (int) _HiddenL)).Format ("\t\t\tHidden Layer: {0:d}".Append( string. _ITerr)).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 00000117 00000118 00000119 string.ReportInterval)).Format ("\t\tMaximum Epoch: {0:d}".IterMax)).Format ("\t\t\tTarget Error: {0:0.Format ("\nReport every {0:d} generations. } public override bool Train() { Page 155 of 225 . base. base.TargetError)).

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 00000146 00000147 00000148 00000149 bool success = true. "Error". Report(BPnet.Show("Error running Backprop training!". } public bool getGraphDataFromDll() { bool success = true. i++) { base. } catch { MessageBox.OK. try { success = BPnet. success = false.besterrorlist. MessageBoxIcon. try { double[] besterrlist = BPnet.Error). for (int i = 0.BestErrList.Add((double)i.Train(). MessageBoxButtons. } success = getGraphDataFromDll(). i < besterrlist. return success._tr_output)._Trained = success. besterrlist[i]). base.Length. Page 156 of 225 .

"Error". MessageBoxIcon. MessageBoxButtons. MessageBoxButtons. } base.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 00000168 00000169 00000170 00000171 00000172 00000173 00000174 00000175 00000176 00000177 00000178 00000179 } } catch { MessageBox. Report(BPnet. } catch { MessageBox. return success. "Error".Show("Error getting graph data from DLL".Error).Test(). } public override bool Test() { bool success = true._te_output).OK.Show("Error running Backprop testing!".Error). success = false. try { success = BPnet.OK. success = false. } return success. MessageBoxIcon. } #endregion Page 157 of 225 ._Tested = success.

protected int _numoutputs. protected double[][][] _weights.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000180 00000181 } } 9.Forms. protected int _numinputs.Model { public class Network { #region data /********************\ |* DATA *| \********************/ //[layer][node][input] protected double[][][] _inputs.3. protected double[][] _outputs. _05025397. System. protected int _hiddennodes. System.Generic.Collections.1. namespace _05025397.“Network. System.1 .Model 9. System.cs” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 using using using using using using System.Windows.Text. Page 158 of 225 .Linq.3 .1.

_weights[1] = new double[_hiddennodes][].William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000025 00000026 00000027 00000028 00000029 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 protected Random Rand. i++) { _inputs[1][i] = new double[_numinputs + 1]. i++) { _inputs[0][i] = new double[2]. i < _numinputs. #endregion #region init /********************\ |* INIT *| \********************/ private void init_inputlayer() { _inputs[0] = new double[_numinputs][]. _weights[0][i] = new double[2]. for (int i = 0. for (int i = 0. } } private void init_hiddenlayer() { _inputs[1] = new double[_hiddennodes][]. _weights[1][i] = new double[_numinputs + 1]. _outputs[1] = new double[_hiddennodes]. i < _hiddennodes. _weights[0] = new double[_numinputs][]. Page 159 of 225 . _outputs[0] = new double[_numinputs].

i++) { _inputs[1][i][_numinputs] = 1.0. _weights[0][i][1] = 1. _weights[2] = new double[_numoutputs][]. i < _numinputs. i < _numoutputs.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000055 00000056 00000057 00000058 00000059 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 } } private void init_outputlayer() { _inputs[2] = new double[_numoutputs][]. for (int i = 0. Page 160 of 225 . _weights[1][i][_numinputs] = 1.0. i++) { _inputs[2][i] = new double[_hiddennodes + 1]. _weights[2][i] = new double[_hiddennodes + 1]. i++) { _inputs[0][i][1] = 1. i < _hiddennodes. _outputs[2] = new double[_numoutputs]. } } private void init_bias() { //Input bias for (int i = 0. } //Hidden layer bias for (int i = 0.0.0.

i < _numoutputs.NextDouble(). j++) { _weights[1][i][j] = Rand. i++) { for (int j = 0.0. i < _hiddennodes.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000085 00000086 00000087 00000088 00000089 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 } //Output layer bias for (int i = 0. } } private void init_hiddenweights() { for (int i = 0. _weights[2][i][_hiddennodes] = 1. i++) { _inputs[2][i][_hiddennodes] = 1. i < _numinputs. j < _numinputs.0. i++) { _weights[0][i][0] = Rand. } } private void init_inputweights() { for (int i = 0.NextDouble(). } } } private void init_outputweights() Page 161 of 225 .

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000115 00000116 00000117 00000118 00000119 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 { for (int i = 0.0.Pow(Math. double sum = 0. j++) { _weights[2][i][j] = Rand. i < _numinputs. try { //Calculate results for this layer for (int i = 0.E. } } } #endregion #region internal /********************\ |* INTERNAL *| \********************/ private double sigmoid(double input) { return (1 / (1 + Math. -input))). j < _hiddennodes. i++) { for (int j = 0. i++) { Page 162 of 225 . } private bool Runinput() { bool success = true.NextDouble(). i < _numoutputs.

MessageBoxIcon. j++) { _inputs[1][i][j] = _outputs[0][j]. Page 163 of 225 .Show("Error processing input layer". i++) { for (int j = 0. i < _hiddennodes.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000145 00000146 00000147 00000148 00000149 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 00000168 00000169 00000170 00000171 00000172 00000173 00000174 for (int j = 0. } return success. } } catch { MessageBox. j++) { sum += (_inputs[0][i][j] * _weights[0][i][j]).0.OK. double sum = 0. j < 2.0. sum = 0. "Error".Error). } private bool Runhidden() { bool success = true. } _outputs[0][i] = sigmoid(sum). j < _numinputs. success = false. MessageBoxButtons. try { //Feed forward the results from input layer for (int i = 0.

MessageBoxIcon.0. MessageBoxButtons. i < _hiddennodes. } _outputs[1][i] = sigmoid(sum). i++) { for (int j = 0. sum = 0. } return success.OK. "Error". j++) { sum += (_inputs[1][i][j] * _weights[1][i][j]). j < _numinputs + 1.0. double sum = 0.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000175 00000176 00000177 00000178 00000179 00000180 00000181 00000182 00000183 00000184 00000185 00000186 00000187 00000188 00000189 00000190 00000191 00000192 00000193 00000194 00000195 00000196 00000197 00000198 00000199 00000200 00000201 00000202 00000203 00000204 } } //Calculate results for this layer for (int i = 0.Error). try Page 164 of 225 .Show("Error processing hidden layer". } } catch { MessageBox. } private bool Runoutput() { bool success = true. success = false.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000205 00000206 00000207 00000208 00000209 00000210 00000211 00000212 00000213 00000214 00000215 00000216 00000217 00000218 00000219 00000220 00000221 00000222 00000223 00000224 00000225 00000226 00000227 00000228 00000229 00000230 00000231 00000232 00000233 00000234 { //Feed forward the results from hidden layer for (int i = 0. success = false. j++) { sum += (_inputs[2][i][j] * _weights[2][i][j]). i++) { for (int j = 0. j++) { _inputs[2][i][j] = _outputs[1][j]. i < _numoutputs.Show("Error processing output layer".Error). i < _numoutputs. MessageBoxIcon. j < _hiddennodes + 1. } return success.OK.0. j < _hiddennodes. } Page 165 of 225 . } _outputs[2][i] = sigmoid(sum). sum = 0. "Error". MessageBoxButtons. } } //Calculate results for this layer for (int i = 0. i++) { for (int j = 0. } } catch { MessageBox.

i++) //Each input node has only one real input //and a bias . MessageBoxIcon. "Incorrect number of inputs". MessageBoxButtons. _inputs[0][i][0] = inputs[i]. if(success) success = Runhidden(). } if (success) for (int i = 0.Error). Page 166 of 225 .OK. i < _numinputs.the number of nodes corresponds //to the number of inputs accepted. if(success) success = Runinput().Length != _numinputs) { MessageBox. //The numbers of inputs must match up if (inputs.Show("Error: Incorrect number of inputs supplied to NN". success = false.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000235 00000236 00000237 00000238 00000239 00000240 00000241 00000242 00000243 00000244 00000245 00000246 00000247 00000248 00000249 00000250 00000251 00000252 00000253 00000254 00000255 00000256 00000257 00000258 00000259 00000260 00000261 00000262 00000263 00000264 #endregion #region external /********************\ |* EXTERNAL *| \********************/ public bool Run(double[] inputs) { bool success = true.

_hiddennodes = hiddennodes. } #endregion #region constructor /********************\ |* CONSTRUCTOR *| \********************/ public Network(int numinputs. return success. _numoutputs = numoutputs.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000265 00000266 00000267 00000268 00000269 00000270 00000271 00000272 00000273 00000274 00000275 00000276 00000277 00000278 00000279 00000280 00000281 00000282 00000283 00000284 00000285 00000286 00000287 00000288 00000289 00000290 00000291 00000292 00000293 00000294 if(success) success = Runoutput(). int hiddennodes. //Set network structure descriptors _numinputs = numinputs. } public double[] GetOutput() { //Return the outputs from the //output layer return _outputs[2]. //We'll always have 3 layers Page 167 of 225 . int numoutputs. int SetRandom) { //Set random number generator Rand = new Random(SetRandom).

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3
00000295 00000296 00000297 00000298 00000299 00000300 00000301 00000302 00000303 00000304 00000305 00000306 00000307 00000308 00000309 00000310 } //Input, Hidden, and Output _inputs = new double[3][][]; _weights = new double[3][][]; _outputs = new double[3][]; init_inputlayer(); init_hiddenlayer(); init_outputlayer(); init_bias(); init_inputweights(); init_hiddenweights(); init_outputweights(); } #endregion }

9.1.3.2 - “GANetwork.cs”
00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 using using using using using using using System; System.Collections.Generic; System.Linq; System.Text; System.Windows.Forms; _05025397; System.Collections;

//This class contains Genetic Algorithm specific functions //in addition to the basic feed foward neural net functionality. namespace _05025397.Model

Page 168 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3
00000013 { 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042

public class GANetwork : Network, IComparable { #region data /********************\ |* DATA *| \********************/ double _MSqE; #endregion #region .NET /********************\ |* .NET *| \********************/ //Interface_Implementation //Return Value Meanings: //-<Zero: x < y //-Zero: x == y //->Zero: x > y public int CompareTo(object a) { GANetwork b; if (a is GANetwork) { b = a as GANetwork; return _MSqE.CompareTo(b.MSqE); }

Page 169 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3
00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 return 0; } #endregion #region external /********************\ |* EXTERNAL *| \********************/ public void getMsQE(double[][] inputs, double[][] outputs) { double sum = 0.0; int counter = 0; double[] netoutputs = new double[_numoutputs]; for (int i = 0; i < inputs.Length; i++) { base.Run(inputs[i]); netoutputs = base.GetOutput(); for (int j = 0; j < netoutputs.Length; j++) { sum += (outputs[i][j] - netoutputs[j]) * (outputs[i][j] - netoutputs[j]); counter++; } } _MSqE = (sum / counter); }

Page 170 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3
00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 public double[] getWeights() { ArrayList collater = new ArrayList(); //Collect all the weights in an array list //then spit them out as a 1D array of doubles try { for (int i = 0; i < _weights.Length; i++) { for (int j = 0; j < _weights[i].Length; j++) { for (int k = 0; k < _weights[i][j].Length; k++) { collater.Add(_weights[i][j][k]); } } } } catch { MessageBox.Show("Fatal Error collating weights to 1D array", "Error", MessageBoxButtons.OK, MessageBoxIcon.Error); Application.Exit(); } return (double[])collater.ToArray(typeof(double)); } public void setWeights(double[] weights)

Page 171 of 225

MessageBoxButtons. j++) { for (int k = 0.Length. k++. i++) { for (int j = 0.Error). j < _weights[i]. i < _weights. "Error".Exit().OK. wc++) { _weights[i][j][k] = weights[wc].Length. wc = 0. } } /********************\ |* GETTERS/SETTERS *| \********************/ public double MSqE { Page 172 of 225 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 00000117 00000118 00000119 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 { //Take a 1D array of doubles and apply //to the correct positions in our weights //array try { for (int i = 0. MessageBoxIcon. } } } } catch { MessageBox. Application.Show("Fatal Error adding 1D weight array to 3D weight array".Length. k < _weights[i][j].

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 00000146 00000147 00000148 00000149 00000150 00000151 } get { return _MSqE. int Nhnodes. } #endregion } Page 173 of 225 . } set { _MSqE = value. Noutputs. } } #endregion #region constructor /********************\ |* CONSTRUCTOR *| \********************/ public GANetwork(int Ninputs. Nhnodes.0. int SetRandom) : base(Ninputs. SetRandom) { MSqE = 0. int Noutputs.

double LearnRate. public: /********************\ |* EXTERNAL *| \********************/ FANN_Cascade(int InputL. using namespace System::Runtime::InteropServices.1.1 . double TargetErr.DLL Wrapper for FANN functions 9. namespace FANN_Wrapper { public ref class FANN_Cascade //Managed wrapper for UnManaged_FANN_CasC //which is a C++ interface to the C programmed //FANN dll. { public: /********************\ |* DATA *| \********************/ UnManaged_FANN_CasC* UMWrapper.h” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 #pragma once #include "UnManaged_FANN_BackProp.4. int OutputL.h" #include "UnManaged_FANN_CasC. Page 174 of 225 .“FANN_Wrapper.4 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9.h" using namespace System.1.

bool Train(void). public: /********************\ |* GETTERS/SETTERS *| \********************/ property String^ _te_output { String^ get() { return gcnew String((wchar_t *) UMWrapper->get_report_test()). } } property array<double>^ besterrorlist { array<double>^ get() MaxNeurons). } } property String^ _tr_output { String^ get() { return gcnew String((wchar_t *) UMWrapper->get_report_train()). bool Test(void). int ~FANN_Cascade(void). !FANN_Cascade(void). Page 175 of 225 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000027 00000028 00000029 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 int ReportInterval.

for(int i = 0. array<double>^ arout = gcnew array<double>(size).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000057 00000058 00000059 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 { int size. } } }. { public: /********************\ |* DATA *| \********************/ UnManaged_FANN_Backprop* UMWrapper. } return arout. i++) { arout[i] = ar[i]. i < size. public ref class FANN_BackProp //Managed wrapper for UnManaged_FANN_Backprop //which is a C++ interface to the C programmed //FANN dll. const double *ar = UMWrapper->get_besterrs(size). public: Page 176 of 225 .

!FANN_BackProp(void). ~FANN_BackProp(void). bool Test(void). } } property String^ _tr_output { String^ get() { return gcnew String((wchar_t *) UMWrapper->get_report_train()).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000087 00000088 00000089 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 /********************\ |* EXTERNAL *| \********************/ FANN_BackProp(int InputL. int ReportInterval. int MaximumIteration). } Page 177 of 225 . int HiddenL. double TargetErr. bool Train(void). public: /********************\ |* GETTERS/SETTERS *| \********************/ property String^ _te_output { String^ get() { return gcnew String((wchar_t *) UMWrapper->get_report_test()). int OutputL. double LearnRate.

array<double>^ arout = gcnew array<double>(size). } return arout. const double *ar = UMWrapper->get_besterrs(size). Page 178 of 225 . for(int i = 0. } } }. i < size.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000117 00000118 00000119 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 } } property array<double>^ besterrorlist { array<double>^ get() { int size. i++) { arout[i] = ar[i].

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3

9.1.4.2 - “FANN_Wrapper.cpp”
00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 #include "stdafx.h" #include "FANN_Wrapper.h" namespace FANN_Wrapper { /********************\ |* FANN_BackProp *| \********************/ //Constructor FANN_BackProp::FANN_BackProp(int InputL, int HiddenL, int OutputL, double LearnRate, double TargetErr, int ReportInterval, int MaximumIteration) { UMWrapper = new UnManaged_FANN_Backprop( InputL, HiddenL, OutputL, LearnRate, TargetErr, ReportInterval, MaximumIteration); } //Destructor FANN_BackProp::~FANN_BackProp(void) { if (UMWrapper) { delete UMWrapper; UMWrapper = 0; } }

Page 179 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3
00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059

//Finalizer FANN_BackProp::!FANN_BackProp(void) { if (UMWrapper) { delete UMWrapper; UMWrapper = 0; } }

//Train bool FANN_BackProp::Train(void) { return UMWrapper->Train(); } //Test bool FANN_BackProp::Test(void) { return UMWrapper->Test(); } /********************\ |* FANN_Cascade *| \********************/ //Constructor FANN_Cascade::FANN_Cascade(int InputL, int OutputL,

Page 180 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3
00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 double LearnRate, double TargetErr, int ReportInterval, int MaxNeurons) { UMWrapper = new UnManaged_FANN_CasC( InputL, OutputL, LearnRate, TargetErr, ReportInterval, MaxNeurons); } //Destructor FANN_Cascade::~FANN_Cascade(void) { if (UMWrapper) { delete UMWrapper; UMWrapper = 0; } } //Finalizer FANN_Cascade::!FANN_Cascade(void) { if (UMWrapper) { delete UMWrapper; UMWrapper = 0; } } //Train

Page 181 of 225

William Sayers

05025397

Genetic Algorithms and Neural Networks
Milestone 3
00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 } bool FANN_Cascade::Train(void) { return UMWrapper->Train(); } //Test bool FANN_Cascade::Test(void) { return UMWrapper->Test(); }

Page 182 of 225

4.“UnManaged_FANN_BackProp.h> class UnManaged_FANN_Backprop //An interface to the C programmed //FANN dll.h” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 #pragma once #include <iostream> #include <string> #include <sstream> #include <vector> #include <doublefann. { private: /********************\ |* DATA *| \********************/ //Network Structure int _InputL.1. int _ReportInterval. //Output for C# Page 183 of 225 . int _OutputL. int _MaximumIteration. //Training Parameters double _LearnRate. double _Momentum.h> #include <fann_cpp. double _TargetErr. int _HiddenL.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9.3 .

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000030 std::wstringstream tr_output. 00000057 00000058 public: 00000059 /********************\ Page 184 of 225 . int OutputL. 00000032 wchar_t *wchoutput. 00000056 bool Test(void). double TargetErr. 00000036 00000037 //FANN Data 00000038 struct fann *ann. 00000033 00000034 std::vector<double> vecbsterr. 00000052 00000053 //Interface functions 00000054 //accessed from C# 00000055 bool Train(void). 00000031 std::wstringstream te_output. 00000040 00000041 public: 00000042 /********************\ 00000043 |* EXTERNAL *| 00000044 \********************/ 00000045 //Constructor 00000046 UnManaged_FANN_Backprop(int InputL. int HiddenL. 00000049 00000050 //Destructor 00000051 ~UnManaged_FANN_Backprop(void). 00000048 int ReportInterval. int MaximumIteration). 00000047 double LearnRate. 00000035 double *bsterr. 00000039 struct fann_train_data *data.

|* GETTERS/SETTERS *| \********************/ const wchar_t* get_report_train(void). Page 185 of 225 . const double* get_besterrs(int &size). const wchar_t* get_report_test(void).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000060 00000061 00000062 00000063 00000064 00000065 00000066 }.

4. } //Destructor UnManaged_FANN_Backprop::~UnManaged_FANN_Backprop() { fann_destroy_train(data). int OutputL. double TargetErr. ann = NULL. _LearnRate = LearnRate.cpp” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 #include "StdAfx.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9.h" #include "UnManaged_FANN_BackProp. _OutputL = OutputL.4 . int MaximumIteration) { _InputL = InputL. _ReportInterval = ReportInterval.1.h" //Constructor UnManaged_FANN_Backprop::UnManaged_FANN_Backprop (int InputL. data = NULL. int ReportInterval.“UnManaged_FANN_BackProp. _MaximumIteration = MaximumIteration. bsterr = NULL. Page 186 of 225 . _HiddenL = HiddenL. wchoutput = NULL. _TargetErr = TargetErr. int HiddenL. double LearnRate.

double error.) { throw("").. bool success = true. ann = fann_create_standard(3. delete wchoutput. } //Train bool UnManaged_FANN_Backprop::Train(void) { static int firstrun = false. success = false. } if (fann_num_input_train_data(data) != _InputL) Page 187 of 225 . wchoutput = NULL.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 delete bsterr. } catch(.. try { data = fann_read_train_from_file("dlltrdata. int reportcounter = 0. _OutputL). bsterr = NULL.dat"). _HiddenL. _InputL.

success = false. //Set the learning rate fann_set_learning_rate(ann. fann_set_activation_function_output (ann. 1). 1). success = false. FANN_TRAIN_BATCH). //Standard backprop fann_set_training_algorithm(ann. //Sigmoid Activation Functions (the same one //the GA uses). } fann_set_activation_steepness_hidden(ann. //Same range the GA's weights are //initialised too Page 188 of 225 . fann_set_activation_steepness_output(ann. fann_set_activation_function_hidden (ann. FANN_SIGMOID). FANN_SIGMOID).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 { throw(""). } if (fann_num_output_train_data(data) != _OutputL) { throw(""). (float) _LearnRate).

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000090 00000091 00000092 00000093 00000094 00000095 00000096 00000097 00000098 00000099 00000100 00000101 00000102 00000103 00000104 00000105 00000106 00000107 00000108 00000109 00000110 00000111 00000112 00000113 00000114 00000115 00000116 00000117 00000118 00000119 fann_randomize_weights(ann. vecbsterr.". if (reportcounter == _ReportInterval) { tr_output << "Best error at iteration [ " << i << " ] was " << error << ".\n". 1. i < _MaximumIteration. Page 189 of 225 . } } fann_destroy_train(data). i = _MaximumIteration + 1. } reportcounter++. if (error < _TargetErr) { tr_output << "\nNetwork matching or improving upon target error" << "found at iteration [ " << i << " ]. 0. //Training Loop for (int i = 0.push_back(error). reportcounter = 0.0. data).0). i++) { //Train one epoch then check the //mean squared error error = (double) fann_train_epoch(ann.

j < _InputL. for (unsigned int i = 0. double *error = NULL. j++) { Page 190 of 225 . } //Test bool UnManaged_FANN_Backprop::Test(void) { fann_type *calc_out = NULL.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000120 00000121 00000122 00000123 00000124 00000125 00000126 00000127 00000128 00000129 00000130 00000131 00000132 00000133 00000134 00000135 00000136 00000137 00000138 00000139 00000140 00000141 00000142 00000143 00000144 00000145 00000146 00000147 00000148 00000149 firstrun = true. } te_output << "\nOutput achieved was". j < _OutputL. error = new double[fann_length_train_data(data)]. fann_scale_train_data(data. data->input[i]). i++) { calc_out = fann_run(ann.dat"). for (int j = 0. i < fann_length_train_data(data). te_output << "\nWith inputs". j++) { te_output << " [" << data->input[i][j] << "] ". data = fann_read_train_from_file("dlltedata. for (int j = 0. 0. return true. 1). te_output << "\n\n\t\t\t~~Network Testing~~".

j++) { te_output << " [" << data->output[i][j] << "] ".str(). wchoutput = NULL.c_str()). te_output << fann_test_data(ann. for (int j = 0. } wchoutput = new wchar_t[tr_output. wcscpy(wchoutput. j < _OutputL.str(). tr_output. delete error. return true.length() + 1]. Page 191 of 225 . data). delete calc_out. } te_output << "\nOutput Desired was". } } te_output << "\nMean Squared Error with these inputs and outputs is:".William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 00000168 00000169 00000170 00000171 00000172 00000173 00000174 00000175 00000176 00000177 00000178 00000179 te_output << " [" << calc_out[j] << "] ". } //get_report const wchar_t* UnManaged_FANN_Backprop::get_report_train(void) { if (wchoutput != NULL) { delete wchoutput.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000180 00000181 00000182 00000183 00000184 00000185 00000186 00000187 00000188 00000189 00000190 00000191 00000192 00000193 00000194 00000195 00000196 00000197 00000198 00000199 00000200 00000201 00000202 00000203 00000204 00000205 00000206 00000207 00000208 00000209 return wchoutput. size = vecbsterr. } const wchar_t* UnManaged_FANN_Backprop::get_report_test(void) { if (wchoutput != NULL) { delete wchoutput. } bsterr = new double[vecbsterr.length() + 1]. } //get_besterrs const double* UnManaged_FANN_Backprop::get_besterrs(int &size) { if (bsterr == NULL) { delete bsterr.size()]. } wchoutput = new wchar_t[te_output. bsterr = NULL. te_output. wcscpy(wchoutput. Page 192 of 225 .c_str()). wchoutput = NULL.size(). return wchoutput.str().str().

int _OutputL.h> class UnManaged_FANN_CasC { public: /********************\ |* DATA *| \********************/ //Network Structure int _InputL. return bsterr.5 . i++) { bsterr[i] = vecbsterr[i]. 9.1.h” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 #pragma once #include <iostream> #include <string> #include <sstream> #include <vector> #include <doublefann.clear(). i < vecbsterr. } vecbsterr. //Training Parameters Page 193 of 225 .4.“UnManaged_FANN_CasC.size().William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000210 00000211 00000212 00000213 00000214 00000215 00000216 00000217 00000218 00000219 } for (unsigned int i = 0.

00000027 wchar_t *wchoutput. 00000047 double LearnRate. 00000043 00000044 //Constructor 00000045 UnManaged_FANN_CasC 00000046 (int InputL. 00000030 double *bsterr. 00000021 int _ReportInterval. Page 194 of 225 . 00000031 00000032 //FANN Data 00000033 struct fann *ann. 00000026 std::wstringstream te_output. 00000022 int _MaxNeurons.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000019 double _LearnRate. double TargetErr. 00000035 00000036 public: 00000037 /********************\ 00000038 |* EXTERNAL *| 00000039 \********************/ 00000040 00000041 //Destructor 00000042 ~UnManaged_FANN_CasC(void). int MaxNeurons). int OutputL. 00000028 00000029 static std::vector<double> vecbsterr. 00000023 00000024 //Output for C# 00000025 static std::wstringstream tr_output. 00000048 int ReportInterval. 00000020 double _TargetErr. 00000034 struct fann_train_data *data.

unsigned int epochs_between_reports. 00000058 float desired_error. Page 195 of 225 . unsigned int epochs). 00000068 }. 00000054 00000055 static int FANN_API Report_Callback 00000056 (struct fann *ann. 00000057 unsigned int max_epochs. 00000053 bool Test(void). 00000066 00000067 const double* get_besterrs(int &size).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000049 00000050 //Interface functions 00000051 //accessed from C# 00000052 bool Train(void). struct fann_train_data *train. 00000059 00000060 public: 00000061 /********************\ 00000062 |* GETTERS/SETTERS *| 00000063 \********************/ 00000064 const wchar_t* get_report_train(void). 00000065 const wchar_t* get_report_test(void).

wchoutput = NULL. _TargetErr = TargetErr. double TargetErr. data = NULL. std::vector<double> UnManaged_FANN_CasC::vecbsterr. _MaxNeurons = MaxNeurons.6 . //Constructor UnManaged_FANN_CasC::UnManaged_FANN_CasC (int InputL. bsterr = NULL.4. _ReportInterval = ReportInterval. int MaxNeurons) { _InputL = InputL.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9. double LearnRate.1. _LearnRate = LearnRate.cpp” 00000001 00000002 00000003 00000004 00000005 00000006 00000007 00000008 00000009 00000010 00000011 00000012 00000013 00000014 00000015 00000016 00000017 00000018 00000019 00000020 00000021 00000022 00000023 00000024 00000025 00000026 00000027 00000028 00000029 #include "StdAfx. ann = NULL. _OutputL = OutputL.“UnManaged_FANN_CasC.h" #include "UnManaged_FANN_CasC. int OutputL. int ReportInterval.h" //Static variable declarations std::wstringstream UnManaged_FANN_CasC::tr_output. } //Destructor UnManaged_FANN_CasC::~UnManaged_FANN_CasC(void) Page 196 of 225 .

c_str()).length() + 1]. return wchoutput. Page 197 of 225 . } const wchar_t* UnManaged_FANN_CasC::get_report_test(void) { if (wchoutput != NULL) { delete wchoutput. fann_destroy_train(data). delete wchoutput.str(). wchoutput = NULL. } wchoutput = new wchar_t[tr_output.str(). wchoutput = NULL.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000030 00000031 00000032 00000033 00000034 00000035 00000036 00000037 00000038 00000039 00000040 00000041 00000042 00000043 00000044 00000045 00000046 00000047 00000048 00000049 00000050 00000051 00000052 00000053 00000054 00000055 00000056 00000057 00000058 00000059 { delete bsterr. wcscpy(wchoutput. } //get_report const wchar_t* UnManaged_FANN_CasC::get_report_train(void) { if (wchoutput != NULL) { delete wchoutput. tr_output. bsterr = NULL.

clear().size().str(). wcscpy(wchoutput.length() + 1]. } Page 198 of 225 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000060 00000061 00000062 00000063 00000064 00000065 00000066 00000067 00000068 00000069 00000070 00000071 00000072 00000073 00000074 00000075 00000076 00000077 00000078 00000079 00000080 00000081 00000082 00000083 00000084 00000085 00000086 00000087 00000088 00000089 wchoutput = NULL. return wchoutput.c_str()). } wchoutput = new wchar_t[te_output. } //get_besterrs const double* UnManaged_FANN_CasC::get_besterrs(int &size) { if (bsterr == NULL) { delete bsterr.str(). } vecbsterr. te_output. for (unsigned int i = 0.size(). size = vecbsterr. i++) { bsterr[i] = vecbsterr[i]. bsterr = NULL. return bsterr. i < vecbsterr. } bsterr = new double[vecbsterr.size()].

00000113 00000114 //Some more network customisation here 00000115 //might be nice in future. Page 199 of 225 .. 00000104 } 00000105 00000106 if (fann_num_input_train_data(data) != _InputL) 00000107 throw(""). 00000118 00000119 fann_set_quickprop_mu(ann. 00000108 00000109 if (fann_num_output_train_data(data) != _OutputL) 00000110 throw("").0).) 00000102 { 00000103 throw(""). 2. 00000096 00000097 try 00000098 { 00000099 data = fann_read_train_from_file("dlltrdata. 0. (float)_LearnRate). 00000094 00000095 ann = fann_create_shortcut(2.0). _InputL. 00000116 00000117 fann_set_quickprop_decay(ann. 00000100 } 00000101 catch(. 00000111 00000112 fann_set_learning_rate(ann.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000090 //Train 00000091 bool UnManaged_FANN_CasC::Train() 00000092 { 00000093 int reportcounter = 0. _OutputL).dat")..

_ReportInterval. 00000137 fann_set_cascade_output_change_fraction(ann. 00000138 fann_set_cascade_candidate_change_fraction(ann. 1. 00000144 00000145 return true. FANN_STOPFUNC_MSE). 0. (float) _LearnRate). 00000146 } 00000147 00000148 //Callback for reporting 00000149 int FANN_API UnManaged_FANN_CasC::Report_Callback Page 200 of 225 . 1). FANN_SIGMOID). 00000124 00000125 fann_set_activation_steepness_output(ann. 00000122 00000123 fann_set_cascade_max_out_epochs(ann. 0. (float) _TargetErr).01f). 00000133 00000134 fann_set_activation_function_output(ann. 00000139 fann_set_train_stop_function(ann. FANN_TRAIN_QUICKPROP). 00000126 fann_set_activation_steepness_hidden(ann. 00000142 00000143 fann_cascadetrain_on_data(ann. 00000129 00000130 fann_set_training_algorithm(ann. UnManaged_FANN_CasC::Report_Callback). 00000135 00000136 fann_randomize_weights(ann.0.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000120 00000121 fann_set_cascade_weight_multiplier(ann. 1). 00000140 00000141 fann_set_callback(ann. 150). FANN_SIGMOID). 00000131 00000132 fann_set_activation_function_hidden(ann. _MaxNeurons.0). 00000127 00000128 fann_set_learning_rate(ann. 1). 0.01f). data.

Page 201 of 225 . if (MSqE <= desired_error) { tr_output << "\nNetwork matching or improving upon target error" << "found at node [ " << node << " ]. vecbsterr.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000150 00000151 00000152 00000153 00000154 00000155 00000156 00000157 00000158 00000159 00000160 00000161 00000162 00000163 00000164 00000165 00000166 00000167 00000168 00000169 00000170 00000171 00000172 00000173 00000174 00000175 00000176 00000177 00000178 00000179 (struct fann *ann.push_back(MSqE). unsigned int max_epochs. fann_scale_train_data(data. unsigned int epochs) { double MSqE = fann_get_MSE(ann). 1). struct fann_train_data *train. } node++.". tr_output << "Best error at node [ " << node << " ] was " << MSqE << ". double *error = NULL. unsigned int epochs_between_reports.\n".dat"). } //Test bool UnManaged_FANN_CasC::Test() { fann_type *calc_out = NULL. data = fann_read_train_from_file("dlltedata. float desired_error. 0. static int node = 1. return 0.

te_output << "\n\n\t\t\t~~Network Testing~~". j++) { te_output << " [" << data->input[i][j] << "] ". Page 202 of 225 . error = new double[fann_length_train_data(data)]. for (int j = 0. data->input[i]). j++) { te_output << " [" << calc_out[j] << "] ". te_output << "\nWith inputs:". NULL). } te_output << "\nOutput achieved was:".William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000180 00000181 00000182 00000183 00000184 00000185 00000186 00000187 00000188 00000189 00000190 00000191 00000192 00000193 00000194 00000195 00000196 00000197 00000198 00000199 00000200 00000201 00000202 00000203 00000204 00000205 00000206 00000207 00000208 00000209 fann_set_callback(ann. } } te_output << "\nMean Squared Error with these inputs and outputs is:". for (int j = 0. j < _InputL. j < _OutputL. for (unsigned int i = 0. i++) { calc_out = fann_run(ann. i < fann_length_train_data(data). j < _OutputL. for (int j = 0. te_output << fann_test_data(ann. j++) { te_output << " [" << data->output[i][j] << "] ". data). } te_output << "\nOutput Desired was:".

delete calc_out.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 00000210 00000211 00000212 00000213 00000214 00000215 } delete error. return true. Page 203 of 225 .

Page 204 of 225 .2 .UML Class Diagram The class diagram on the following pages represents the structure of the various classes within the program and how they are combined together to create a whole.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9.

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Page 205 of 225 .

William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Page 206 of 225 .

mit.3 .edu/ga/ o Accessed: 31/10/2008 Page 207 of 225 .William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 -BackProp Previous Page -CascadeCorrelation 1 FANN_Cascade -UMWrapper +FANN_Cascade(in AllDataToFillAttributes) +Train() : bool +Test() : bool 1 -UMWrapper 1 FANN_BackProp +FANN_BackProp(in AllDataToFillAttributes) +Train() : bool +Test() : bool 1 1 1 UnManaged_FANN_BackProp -_InputL : int -_HiddenL : int -_OutputL : int -_LearnRate : double -_TargetErr : double -_ReportInterval : int -_MaximumIteration : int -_Momentum : double -tr_output -te_output -wchoutput -vecbsterr -bsterr -ann -data +UnManaged_FANN_Backprop(in AddDataToFillAttributes) +~UnManaged_FANN_Backprop() +Train() : bool +Test() : bool UnManaged_FANN_CasC -_InputL : int -_OutputL : int -_LearnRate : double -_TargetErr : double -_ReportInterval : int -_MaxNeurons : int -tr_output -te_output -wchoutput -vecbsterr -bsterr -ann -data +UnManaged_FANN_CasC(in AllDataToAFillAttributes) +~UnManaged_FANN_CasC() +Train() : bool +Test() : bool +Report_Callback(in Attributes) : int 1 1 1 1 FANN_Library 9.Libraries Researched • GALib o Matthew Wall o http://lancet.

edu/au/~daniel/software/libneural/ o Accessed: 31/10/2008 • Libann o Anonymous o http://www.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 • Libneural o D.html o Accessed: 31/10/2008 • Flood o Roberto Lopez o http://www.org/libann/index.cimne. Franklin o http://ieee.nongnu.net/ o Accessed: 31/10/2008 • FANN (Fast Artificial Neural Network Library) o Steffen Nissen Page 208 of 225 .uow.sourceforge.com/flood/ o Accessed: 31/10/2008 • Annie o Asim Shankar o http://annie.

Datasets • XOR 1 0 1 0 1 1 1 1 0 0 0 0 • • Fishers Iris Data o http://www.uk/pub/PRNN/README.edu/stat/data/Fisher.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 o http://leenissen.4 .html Page 209 of 225 .xhtml Virus Classification o http://www.stats.ox.math.uah.ac.dk/fann/index.php o Accessed: 31/10/2008 9.

Page 210 of 225 . and in which cases these methods are suitable. and the conclusions derived from that evaluation. Identify the most effective ways of training neural networks using genetic algorithms.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 9.Objectives FINAL YEAR DEGREE PROJECT Project Objectives & Deliverables • • • • • • Research the fields of genetic algorithms and neural networks. Deliverable two: An application meeting the requirements of objectives three and four. evaluate training methods for neural networks. Deliverable three: A document detailing and explaining the evaluation of different training techniques. Build an application that is capable of training a neural network with a genetic algorithm and with a backpropagation system. Using above application. Build a user-interface for the aforementioned application allowing for easy alteration of appropriate variables.5 . and a document describing its implementation and usage. • • • Deliverable one: A research document describing and detailing the research done for objectives one and two. Research current information on training neural networks using genetic algorithms. as well as previous research. and that allows manual adjustment of relevant variables.

G.pdf Fahlman. Retrieved April 16. 2009. January 16). Carnegie Mellon: http://www2. E.unikarlsruhe. (1992). S. C.cmu. Curved trajectory prediction using a self-organizing neural network. 2009. M. 2009. (2007.aifb. Retrieved April 16. S. A. from http://www. Chapter 14: Neural Networks. London: John Murray. E. M..cs. (2004).cmu. S. or the preservation of favoured races in the struggle for life. & Lebiere. J.ps.com/KB/recipes/NeuralNetwork_1. (1988). Retrieved April 13.aspx Page 211 of 225 .edu/afs/cs.edu/user/sef/www/publications/qp-tr. 2009. Davis.edu/afs/cs.cmu. The Cascade-Correlation Learning Architecture.codeproject. 2009.ps Lysle. D. AI for Game Developers (pp.cmu. & G. (1859).aspx Blum. Branke. & Seeman.cs. In D. C. Passing Data between Windows Forms.csharpcorner. Bourg. Retrieved April 13.edu/~chris2d/papers/CED_thesis. Seeman. Bourg. from School of Computer Science. C. Retrieved April 13. from http://www. Graphics processing unit computation of neural networks. Wiley Professional Computing.gz Darwin. (2006). 356-358). On the origin of the species by means of natural selection. 2009. from Codeproject: www. (1991).de/~jbr/Papers/GaNNOverview. AI: Neural Network for Beginners pt. E.edu/user/sef/www/publications/cascor-tr.ps Fahlman.unm. Retrieved April 16.. Carnegie Mellon: http://www-2. S. An Empirical Study of Learning Speed in Back-Propagation Networks.Works Cited Barber. O'Reilly. from School of Computer Science. 1. (1995).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 10 . Neural Network in C++.cs. (2001).com/UploadFile/scottlysle/PassData01142007013005AM/PassData. from C# Corner: http://www.

2009. (2009).William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Marczyk. Microsoft. S.com/en-us/library/system.talkorigins. Retrieved April 15.cs. A. W. The perceptron: A probabilistic theory for information storage and organization in the brain. & Pitts.com/en-us/library/2x8kf7zx.microsoft. (n. (2008).htm Rosenblatt. from http://www. Invention Report S81-64. 2009.gz Matthews. & Srikanth. Retrieved 21 01. Learning Logic. Office of Technology Licensing.backgroundworker.unc. M.componentmodel. W. J. from generation5: http://www. from MSDN: http://msdn. Curved trajectory prediction using a selforganizing neural network. from MSDN: http://msdn. 2009. (1996). Page 212 of 225 . Using C++ Interop (Implicit PInvoke). A.edu/~marshall/WWW/PAPERS/curve9912. 65. Retrieved April 16.asp McCulloch. Retrieved April 15. 2009. Retrieved April 16.ps. from Artificial Neural Networks: http://www. (1982). Bulletin of Mathematical Biophysics .d.microsoft.. Genetic algorithms and evolutionary computation.learnartificialneuralnetworks. Back-propagation for the Uninitiated.generation5. (1943).. F. Retrieved April 13.aspx Microsoft. Stanford University. Robot Control. 386-408. from Computer Science . 2009. (2004).UNC-Chapel Hill: http://www. 5.com/faqs/genalg/genalg.com/robotcontrol. 115-133. V.org/content/2002/bp. (1958).dk/fann/ Parker. A logical calculus of the ideas immanent in nervous activity. Psychological Review . (1999).html Marshall. Nissen. Massachusetts: MIT Press.). B. BackgroundWorker Class. J. 2009. (2009). (2002). from FANN: http://leenissen. An introduction to genetic algorithms. 2009.aspx Mitchell. D. Retrieved April 15.

(n. (1974). Harvard University. P.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Stergiou. 2009. Retrieved 21 01. from Imperial College London: http://www.doc.uk/~nd/surprise_96/journal/vol4/cs11/report. & Dimitrios.html#Contents Werbos. Beyond Regression: New tools for prediction and analysis in the behavioral sciences.ic. (1997). PhD.).org/wiki/index.ac.. Neural Networks. S. C. Retrieved April 15. ZedGraph.d. from http://zedgraph.php?title=Main_Page Page 213 of 225 . 2009.

C. (2004). Wiley Professional Computing. D. AI for Game Developers (pp. from Codeguru: http://www. Neural Network in C++. Retrieved April 13. 3. & G. S. Seeman. M. (2001). Curved trajectory prediction using a self-organizing neural network.ps. (2002).gz Darwin. Bourg.aspx Blum.pdf DiLascia. 1.unikarlsruhe. (1995).codeproject. On the origin of the species by means of natural selection. 2009. Branke. 2009. London: John Murray. A.aspx Barber.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 11 . 2009. Retrieved April 13.codeproject. 2009. (2006). from Codeproject: www. 2009.cs.com/KB/recipes/NeuralNetwork_1.unm. (1992). August 6). & Seeman. from MSDN: http://msdn. 356-358).de/~jbr/Papers/GaNNOverview. from Codeproject: http://www. Retrieved April 13. from http://www.aifb. Retrieved April 13.com/Cpp/Cpp/cpp_mfc/tutorials/article.microsoft. AI: Neural Network for Beginners pt. 2009.com/KB/cs/GA_ANN_XOR. 2. M.aspx Barber. C. Retrieved April 13. or the preservation of favoured races in the struggle for life. Graphics processing unit computation of neural networks.. AI: Neural Network for Beginners pt. E. from Codeproject: http://www.Bibliography Barber. (2007. AI: Neural Network for Beginners pt. G.codeguru.com/KB/recipes/Backprop_ANN. P. Retrieved April 13. Davis. In D. (2006). 2009. J.com/enus/magazine/cc301501. from http://www. Bourg.php/c9855/ Page 214 of 225 .aspx DLL Tutorial for Beginners. S. S. Retrieved April 13. Killing Processes Cleanly. (2006). Chapter 14: Neural Networks. (1859). O'Reilly. Call Unmanaged DLLs from C#.codeproject.edu/~chris2d/papers/CED_thesis.

generation5.aspx Marczyk. W.unc.org/content/2000/xorga. 2009. 2009. J.google. Retrieved April 15.d.ps GoogleChartSharp.d. Carnegie Mellon: http://www2.asp Matthews. January 16).cs.html Marshall. J. 2009.edu/afs/cs. A.cmu.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Fahlman. (1943).cmu. 115-133. (2004). (2002). 2009.cmu. 2009.asp Matthews.generation5. Retrieved April 16. 2009. The Cascade-Correlation Learning Architecture. (1991). from XORGA: http://www. (2000).ps Fahlman. E. Back-propagation for the Uninitiated.org/content/2000/cbpnet. from C# Corner: http://www. from http://code. Retrieved April 13. E.com/UploadFile/scottlysle/PassData01142007013005AM/PassData. Retrieved April 15. & Pitts.generation5. Back-Propagation: CBPNet.cmu. S.edu/afs/cs..com/faqs/genalg/genalg. Page 215 of 225 . 5. Retrieved 26 01. S. Passing Data between Windows Forms. Bulletin of Mathematical Biophysics . Carnegie Mellon: http://www-2. & Srikanth. Genetic algorithms and evolutionary computation.cs.). from Computer Science .edu/~marshall/WWW/PAPERS/curve9912. An Empirical Study of Learning Speed in Back-Propagation Networks. from School of Computer Science.talkorigins. Retrieved April 15.). V. (1988). (n. Retrieved 21 01.cs. (1999). A logical calculus of the ideas immanent in nervous activity. from School of Computer Science.asp McCulloch.generation5. W.ps. C. from http://www.org/content/2002/bp. (2007. 2009. from generation5: http://www. A.UNC-Chapel Hill: http://www.edu/user/sef/www/publications/qp-tr.com/p/googlechartsharp/ Lysle. 2009.csharpcorner.. & Lebiere.gz Matthews. (n. S. J.. J. 2009. Retrieved April 16. Retrieved April 16. from generation5: http://www. Curved trajectory prediction using a selforganizing neural network.edu/user/sef/www/publications/cascor-tr.

(2008). 1990 IJCNN International Joint Conference on .). 2009. Retrieved April 16. Retrieved 21 01. BackgroundWorker Class.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Microsoft. Retrieved 21 01. (1990). Retrieved 21 01. Nissen. Learning Logic. Invention Report S81-64. (n. S. S.d.. S. 2009.dk/fann/html/files/fann_data-h. Reference Manual.componentmodel. Retrieved 21 01.d. 2009.com/en-us/library/2x8kf7zx. (n. S. from FANN: http://leenissen. Neural Networks. (n.html Nissen. from Datatypes: http://leenissen.htm Page 216 of 225 . 2009.microsoft. 2009.microsoft. 2009. Massachusetts: MIT Press. (n.d. (n.aspx Microsoft. (1982). 2009.dk/fann/ Nissen. B.d. Retrieved 21 01. from FANN: http://leenissen.).. (2009). struct fann.html Nplot Charting Library for . from FANN: http://leenissen.backgroundworker. Retrieved 26 01.). from MSDN: http://msdn. Nguyen. from FANN: http://leenissen. D. Office of Technology Licensing.). (n. B. Improving the learning speed of 2-layer neural networks by choosinginitial values of the adaptive weights. 1990.dk/fann/fann_1_2_0/r1597. & Widrow. Retrieved April 16. An introduction to genetic algorithms. 21-26. M.).d.com/en-us/library/system. Retrieved April 15.aspx Mitchell. Using C++ Interop (Implicit PInvoke). from http://netcontrols.org/nplot/wiki/ Parker. struct fann_train_data. (2009). 2009. from MSDN: http://msdn.html Nissen.learnartificialneuralnetworks.com/robotcontrol. Stanford University. from Artificial Neural Networks: http://www.dk/fann/fann_1_2_0/r1837.dk/fann/html/files/fann-h. (1996).d.). Robot Control. D. 2009. S.html#fann_train_enum Nissen.NET.

........ PhD. Harvard University... S.... 10 Figure 2 ........ 13 Figure 4 .....................ac.........Table of Figures Figure 1 ... 2009. from Imperial College London: http://www...... Retrieved 21 01..........org/wiki/index..................William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Rosenblatt...................... . The Cascade-Correlation Learning Architecture. 65................ The circular connections are adjustable weights (Fahlman & Lebiere..........The initial state of a cascade correlation neural network.php?title=Main_Page 12 ......html#Contents Werbos........... Psychological Review ...............A human neuron ...................... 1991)........................... F......An artificial neuron ...................... P........ Neural Networks......Tables 12... 12 Figure 3 ......)..1 ...............local optimum (units are undefined in this case)......... 27 Figure 5 ........... 30 Figure 6 ........ 35 Page 217 of 225 ...........uk/~nd/surprise_96/journal/vol4/cs11/report.............. Stergiou.... Retrieved April 15.. from http://zedgraph..... (1974)......The sigmoid function shape ..... (1958).. The perceptron: A probabilistic theory for information storage and organization in the brain....... Beyond Regression: New tools for prediction and analysis in the behavioral sciences..... 386-408... (1997)..... (n........... C........ic.............. ....................................... ZedGraph............ & Dimitrios...................d....A feed-forward neural network ......doc....... 2009....................

...........Dataset selection tab design (Wizard form)........................ 47 Figure 20 .. The Cascade-Correlation Learning Architecture...........Dataset tab design (Main form) ............. 46 Figure 19 .......................................... 46 Figure 18 .Dataset display design (About form) .........................................................Design of the network settings tab (Wizard form) .......... 1991).... A first node has been added....... 51 Page 218 of 225 ........................ 45 Figure 17 .........Working form design (Working form) ........................Network and Output tab............... 50 Figure 23 ..........................Design of the training algorithm tab (Wizard form) ............................................................................William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Figure 7 ............................................... 1991)............ with two nodes added (Fahlman & Lebiere................................ The CascadeCorrelation Learning Architecture... 36 Figure 9 – Network and output tab design (Main form) ................. 48 Figure 21 ..............The third state of a cascade correlation neural network................................................................................................ .................................. 44 Figure 14 .......................... 42 Figure 12 .............. 35 Figure 8 .................. 44 Figure 15 ........................................... 41 Figure 10 .....................Mean squared error graph tab ............................................... 49 Figure 22 ..............Licenses display design (About form) ....Mean squared error tab design (Main form) .............................. 43 Figure 13 ............................................ The square connections are locked weights(Fahlman & Lebiere....................The dataset tab ...........................Learning Algorithm tab ............................................ 45 Figure 16 .........The second state of a cascade correlation neural network..................................About form design (About form) ........... 42 Figure 11 ..............................Network Settings tab.......

......... 64 Figure 34......................... Licenses view ....... 2002) ....About form........................................ 67 Figure 37 ...........................2 ..........................Mean squared error upon last training test (Viruses)..... 51 Figure 25 ............ 65 Figure 35 ........................................................... 54 Figure 28 .......................About form.............. main view ..........A sigmoid function(Bourg & Seeman..Mean squared error upon last training test (XOR) .................William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Figure 24 ........................ Datasets view ..... In this example “y” is the function output and “t” is the function input............... 61 Figure 31 .....................................................................Number of iterations to achieve target (Iris data) ...........................Number of iterations to achieve target (XOR) ..Dataset tab .............. 68 12............................................Working form ........... 61 Figure 30 ...........Table of Equations Equation 1 ................................ 54 Figure 29 ................................. Back-propagation for the Uninitiated.................About form..................................Number of iterations to achieve target (Viruses) .............................. 52 Figure 26 ........ as defined in the graph above............. 67 Figure 36 .......Mean squared error upon last training test (Iris data) ...........The mean squared error at testing (XOR).......... 64 Figure 33....... 21 Page 219 of 225 ......................................... 62 Figure 32 ..........................................The mean squared error at testing (Iris data) ........... 2004)................ ...................... 53 Figure 27 ..................................................... 12 Equation 2 .......Mean squared error at testing (Viruses)..Summation of the inputs and weights .................. 13 Equation 3 – The delta rule(Matthews.....

........ 2002) ........The definition of 's' given that 'o' is the network output at which the error is measured.................................... Backpropagation for the Uninitiated.............................................. 59 Page 220 of 225 .. Back-propagation for the Uninitiated.......... where ‘oo’ is the sign of the correlation between the candidates value output ‘o’.............................. ................ 21 Equation 6 ... 21 Equation 7 ..differentiated sigmoid function (Matthews............ 2002)......................................... The CascadeCorrelation Learning Architecture...................The statistical approach to deciding relevant data .. 34 12........... 34 Equation 10 ...........Back-propagation testing data (XOR)...........Table of tables Table 1 .Genetic Algorithm testing data (XOR) ... Back-propagation for the Uninitiated.......................................................................................William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Equation 4 .................................... 2002) . 'v' is the candidate unit's value..........................Altered delta calculation rule for the output layer(Matthews.................3 .......................... 1991)........ 2002)....... 15 Table 2 .... 33 Equation 9 ...................... ............................................................... the quantities ‘`v ’ and ‘`Eo’ are the values of v and Eo averaged over all patterns (Fahlman & Lebiere.................................................Altered delta calculation rule for the hidden layer(s) (Matthews..... 1991).......... 'p' is the training pattern............................................Altered perceptron learning rule (Matthews....................................... 21 Equation 5 ...............The partial derivative of ‘s’ with respect to each of the candidates incoming weights ‘wi’......... ‘f'p’ is the derivative for pattern 'p' of the candidate unit's activation function with respect to the sum of it’s inputs and ‘Ii..............................................p’ is the input the candidate receives from unit ‘i’ for pattern ‘p’........ 59 Table 3 ...................The partial derivative of ’s’ (Fahlman & Lebiere.... The Cascade-Correlation Learning Architecture...... 22 Equation 8 .................................................................. Backpropagation for the Uninitiated..............

.................................................... 66 Table 9 ......................Cascade correlation testing data (Iris data) ..................Genetic algorithm testing data (Iris data) .........................................Cascade Correlation testing data (Viruses) ............. 60 Table 5 ................................................................................. 63 Table 7 ..... 63 Table 8 ........ 66 Table 10 ......................William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 Table 4 ............Cascade Correlation testing data (XOR) .............................................Back-propagation testing data (Iris data) ...................Genetic Algorithms testing data (Viruses) ...............Back-propagation testing data (Viruses) ....... 66 Page 221 of 225 ....................... 63 Table 6 .........

o Agreed on further investigation into GALib and genetic algorithms. o Decided on XOR as a starting point. 17/10/2008 o Investigating more deeply into journals based on research already conducted. • 31/10/2008 o First presentation complete and ready o References building up o Project plan complete • 10/11/2008 o Presentation went well o Progress satisfactory • 14/11/2008 o Discussed document formatting and layout Page 222 of 225 . however.Diary • 01/10/2008 – Research progress satisfactory. o Submitted objectives. • 24/10/2008 o University computers were down so code demo was impossible. o Decision made to adjust the weights of the neural network during training as opposed to using a genetic algorithm to establish network structure. • 10/10/2008 o GALib decided as a genetic algorithm starting point o Further non-web references as a goal • • 16/10/2008 o Investigation into various neural network libraries taking place. back-propagation network was running and learning XOR.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 13 .

• • • • • • • • • 19/11/2008 o Milestone one submitted successfully. • • 06/02/2009 o Still exploring extra areas for project to cover. o Progress satisfactory • 20/02/2009 o Progress satisfactory Page 223 of 225 . 13/02/2009 o Have decided to extend the project by including cascade correlation learning algorithm and comparison to genetic algorithms. 21/11/2008 o Progress satisfactory 28/11/2008 o Progress satisfactory 05/12/2008 o Progress satisfactory 12/12/2008 o Progress satisfactory 09/01/2009 o Progress satisfactory 16/01/2009 o Progress satisfactory 23/01/2009 o Progress satisfactory 30/01/2009 o Milestone two has been successfully submitted o Started investigating other areas to explore outside of project requirements.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 o Progress satisfactory.

o Report satisfactory • 05/04/2009 o Further work on user interface needed o Finish C# genetic algorithm code (C++ code does work but is very slow) o Report satisfactory • 06/04/2009 o C# genetic algorithm code is complete. catch-up time for previous two weeks will be used over Easter break. o Have started report • 02/04/2009 o Main problems experienced with genetic algorithm seem related to memory leaks.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 • • • • 27/02/2009 o Progress satisfactory 06/03/2009 o Project progress delayed by other coursework 13/03/2009 o Project progress delayed by other coursework 20/03/2009 o Project progressing. Page 224 of 225 . is much faster (fixed several bugs along the way) and retains none of the problematic memory leaks. have decided to implement in managed language (C#) to try to solve this o Report continues satisfactorily • 04/04/2009 o To do: About box Add dataset notes Decided on virus classification as a third dataset via the custom dataset functionality.

o Report satisfactory.William Sayers 05025397 Genetic Algorithms and Neural Networks Milestone 3 o Report satisfactory • 10/04/2009 o Fully functional project – commencing report. Page 225 of 225 .

Sign up to vote on this title
UsefulNot useful