Data Mining Cluster Analysis: Basic Concepts and Algorithms Lecture Notes for Chapter 8 Introduction to Data Mining

by Tan, Steinbach, Kumar

© Tan,Steinbach, Kumar

Introduction to Data Mining

4/18/2004

1

What is Cluster Analysis?

Finding groups of objects such that the objects in a group will be similar (or related) to one another and different from (or unrelated to) the objects in other groups
Intra-cluster distances are minimized Inter-cluster distances are maximized

© Tan,Steinbach, Kumar

Introduction to Data Mining

4/18/2004

‹#›

Applications of Cluster Analysis

Understanding
– Group related documents for browsing, group genes and proteins that have similar functionality, or group stocks with similar price fluctuations

Discovered Clusters

Industry Group

1 2 3 4

Applied-Matl-DOWN,Bay-Network-Down,3-COM-DOWN, Cabletron-Sys-DOWN,CISCO-DOWN,HP-DOWN, DSC-Comm-DOWN,INTEL-DOWN,LSI-Logic-DOWN, Micron-Tech-DOWN,Texas-Inst-Down,Tellabs-Inc-Down, Natl-Semiconduct-DOWN,Oracl-DOWN,SGI-DOWN, Sun-DOWN Apple-Comp-DOWN,Autodesk-DOWN,DEC-DOWN, ADV-Micro-Device-DOWN,Andrew-Corp-DOWN, Computer-Assoc-DOWN,Circuit-City-DOWN, Compaq-DOWN, EMC-Corp-DOWN, Gen-Inst-DOWN, Motorola-DOWN,Microsoft-DOWN,Scientific-Atl-DOWN Fannie-Mae-DOWN,Fed-Home-Loan-DOWN, MBNA-Corp-DOWN,Morgan-Stanley-DOWN Baker-Hughes-UP,Dresser-Inds-UP,Halliburton-HLD-UP, Louisiana-Land-UP,Phillips-Petro-UP,Unocal-UP, Schlumberger-UP

Technology1-DOWN

Technology2-DOWN

Financial-DOWN Oil-UP

Summarization
– Reduce the size of large data sets
Clustering precipitation in Australia

© Tan,Steinbach, Kumar

Introduction to Data Mining

4/18/2004

‹#›

by last name  Results of a query – Groupings are a result of an external specification  Graph partitioning – Some mutual relevance and synergy.Steinbach. but areas are not identical © Tan.What is not Cluster Analysis?  Supervised classification – Have class label information  Simple segmentation – Dividing students into different registration groups alphabetically. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

Notion of a Cluster can be Ambiguous How many clusters? Six Clusters Two Clusters Four Clusters © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

Steinbach.Types of Clusterings   A clustering is a set of clusters Important distinction between hierarchical and partitional sets of clusters Partitional Clustering – A division data objects into non-overlapping subsets (clusters) such that each data object is in exactly one subset   Hierarchical clustering – A set of nested clusters organized as a hierarchical tree © Tan. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

Kumar Introduction to Data Mining 4/18/2004 ‹#› .Steinbach.Partitional Clustering Original Points A Partitional Clustering © Tan.

Steinbach.Hierarchical Clustering p1 p3 p2 p4 p1 p2 Traditional Hierarchical Clustering p3 p4 Traditional Dendrogram p1 p3 p2 p4 p1 p2 Non-traditional Hierarchical Clustering © Tan. Kumar p3 p4 Non-traditional Dendrogram 4/18/2004 ‹#› Introduction to Data Mining .

Steinbach. we only want to cluster some of the data  Heterogeneous versus homogeneous – Cluster of widely different sizes. – Can represent multiple classes or „border‟ points  Fuzzy versus non-fuzzy – In fuzzy clustering. and densities ‹#› © Tan. points may belong to multiple clusters. a point belongs to every cluster with some weight between 0 and 1 – Weights must sum to 1 – Probabilistic clustering has similar characteristics  Partial versus complete – In some cases. Kumar Introduction to Data Mining 4/18/2004 . shapes.Other Distinctions Between Sets of Clusters  Exclusive versus non-exclusive – In non-exclusive clusterings.

Kumar .Types of Clusters  Well-separated clusters Center-based clusters Contiguous clusters Density-based clusters Property or Conceptual Described by an Objective Function Introduction to Data Mining 4/18/2004 ‹#›      © Tan.Steinbach.

Steinbach. 3 well-separated clusters © Tan.Types of Clusters: Well-Separated  Well-Separated Clusters: – A cluster is a set of points such that any point in a cluster is closer (or more similar) to every other point in the cluster than to any point not in the cluster. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

than to the center of any other cluster – The center of a cluster is often a centroid.Types of Clusters: Center-Based  Center-based – A cluster is a set of objects such that an object in a cluster is closer (more similar) to the “center” of a cluster. Kumar Introduction to Data Mining 4/18/2004 ‹#› . or a medoid. the most “representative” point of a cluster 4 center-based clusters © Tan. the average of all the points in the cluster.Steinbach.

Types of Clusters: Contiguity-Based  Contiguous Cluster (Nearest neighbor or Transitive) – A cluster is a set of points such that a point in a cluster is closer (or more similar) to one or more other points in the cluster than to any point not in the cluster. 8 contiguous clusters © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

6 density-based clusters © Tan.Steinbach. which is separated by low-density regions. – Used when the clusters are irregular or intertwined. Kumar Introduction to Data Mining 4/18/2004 ‹#› . from other regions of high density. and when noise and outliers are present.Types of Clusters: Density-Based  Density-based – A cluster is a dense region of points.

Steinbach. 2 Overlapping Circles © Tan. .Types of Clusters: Conceptual Clusters  Shared Property or Conceptual Clusters – Finds clusters that share some common property or represent a particular concept. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

 Parameters for the model are determined from the data.  © Tan.Steinbach.Types of Clusters: Objective Function  Clusters Defined by an Objective Function – Finds clusters that minimize or maximize an objective function. Mixture models assume that the data is a „mixture' of a number of statistical distributions. (NP Hard) – Can have global or local objectives. Kumar Introduction to Data Mining 4/18/2004 ‹#› . – Enumerate all possible ways of dividing the points into clusters and evaluate the `goodness' of each potential set of clusters by using the given objective function.  Hierarchical clustering algorithms typically have local objectives  Partitional algorithms typically have global objectives – A variation of the global objective function approach is to fit the data to a parameterized model.

Types of Clusters: Objective Function …  Map the clustering problem to a different domain and solve a related problem in that domain – Proximity matrix defines a weighted graph.Steinbach. one for each cluster. and the weighted edges represent the proximities between points – Clustering is equivalent to breaking the graph into connected components. Kumar Introduction to Data Mining 4/18/2004 ‹#› . – Want to minimize the edge weight between clusters and maximize the edge weight within clusters © Tan. where the nodes are the points being clustered.

e. Kumar .g. autocorrelation    Dimensionality Noise and Outliers Type of Distribution Introduction to Data Mining 4/18/2004 ‹#› © Tan..Characteristics of the Input Data Are Important   Type of proximity or density measure – This is a derived measure.Steinbach. but central to clustering Sparseness – Dictates type of similarity – Adds to efficiency  Attribute type – Dictates type of similarity  Type of Data – Dictates type of similarity – Other characteristics.

Kumar Introduction to Data Mining 4/18/2004 ‹#› .Clustering Algorithms    K-means and its variants Hierarchical clustering Density-based clustering © Tan.Steinbach.

must be specified The basic algorithm is very simple © Tan.K-means Clustering      Partitional clustering approach Each cluster is associated with a centroid (center point) Each point is assigned to the cluster with the closest centroid Number of clusters.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› . K.

Most of the convergence happens in the first few iterations. I = number of iterations. K = number of clusters. – Often the stopping condition is changed to „Until relatively few points change clusters‟ n = number of points. – Clusters produced vary from one run to another. etc. cosine similarity. „Closeness‟ is measured by Euclidean distance.     The centroid is (typically) the mean of the points in the cluster. correlation.Steinbach. K-means will converge for common similarity measures mentioned above.K-means Clustering – Details  Initial centroids are often chosen randomly. Kumar . d = number of attributes Introduction to Data Mining 4/18/2004 ‹#›  Complexity is O( n * K * I * d ) – © Tan.

5 2 Original Points 1.5 1.5 2 x x Optimal Clustering © Tan.5 0 0.5 0 0.5 -1 -0.5 1 1.5 0 -2 -1.5 y 1 0. Kumar Introduction to Data Mining Sub-optimal Clustering 4/18/2004 ‹#› .5 y 1 y 1 0.Two different K-means Clusterings 3 2.5 0 0.5 0.5 2 x 3 3 2.5 -1 -0.5 -1 -0.Steinbach.5 2 2 1.5 1 1.5 1 1.5 2 -2 -1.5 2.5 0 0 -2 -1.

Kumar Introduction to Data Mining 4/18/2004 ‹#› .5 2 x © Tan.5 y 1 0.Steinbach.Importance of Choosing Initial Centroids Iteration 1 6 5 4 3 2 3 2.5 -1 -0.5 1 1.5 0 0.5 2 1.5 0 -2 -1.

5 0 0.5 1 1.5 2 -1.5 Iteration 5 3 2.5 1 1.5 0 0.5 1 1.5 0 0.5 1.5 0 0.5 Iteration 6 2 2 2 1.5 0 0.5 2.5 1.5 Iteration 3 2 2 2 1.5 -1 -0.Steinbach.5 2 -2 0.5 0 0 -2 -1.5 0.5 2 x x x Iteration 4 3 3 2.5 -1 -0.5 -1 -0.5 -1 -0.5 0 -2 -1.5 1 1.5 0.5 1 1.5 y y 1 1 y 1 0.5 0 0.Importance of Choosing Initial Centroids Iteration 1 3 3 2.5 0 -2 -1.5 2 -2 0.5 2.5 -1 -0.5 0 0 -2 -1.5 Iteration 2 3 2. Kumar Introduction to Data Mining 4/18/2004 ‹#› .5 1.5 2 x x x © Tan.5 1.5 -1 -0.5 y y 1 1 y 1 0.5 1 1.5 2 -1.

Kumar Introduction to Data Mining 4/18/2004 ‹#› .Evaluating K-means Clusters  Most common measure is Sum of Squared Error (SSE) – For each point.Steinbach. we can choose the one with the smallest error – One easy way to reduce SSE is to increase K. we square these errors and sum them. x) i 1 xCi K – x is a data point in cluster Ci and mi is the representative point for cluster Ci  can show that mi corresponds to the center (mean) of the cluster – Given two clusters. the number of clusters A good clustering with smaller K can have a lower SSE than a poor clustering with higher K  © Tan. SSE   dist 2 (mi . the error is the distance to the nearest cluster – To get SSE.

5 -1 -0.Importance of Choosing Initial Centroids … Iteration 1 5 4 3 2 3 2.5 0 0. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Steinbach.5 0 -2 -1.5 1 1.5 y 1 0.5 2 1.5 2 x © Tan.

5 2.5 0 0 -2 -1.5 1.5 0.5 0 0.5 0 0.5 y 1 y 1 0.5 1 1.5 Iteration 2 2 2 1.5 2 -2 0.5 2 -2 -1.5 2 -1.5 -1 -0.5 1 1.5 -1 -0.5 y y 1 1 y 1 0.5 1 1.5 -1 -0.5 2.5 -1 -0.5 0 0 -2 -1.5 1.5 2 x x Iteration 3 3 3 2.5 -1 -0.5 Iteration 4 3 2.5 0 -2 -1.5 0 0.5 Iteration 5 2 2 2 1.5 0.5 1 1.5 2 x x x © Tan.5 0 0.5 1. Kumar Introduction to Data Mining 4/18/2004 ‹#› .5 1 1.Importance of Choosing Initial Centroids … Iteration 1 3 3 2.5 0 0.Steinbach.

Steinbach. then – – For example.Problems with Selecting Initial Points  If there are K „real‟ clusters then the chance of selecting one centroid from each cluster is small. if K = 10. Kumar Introduction to Data Mining 4/18/2004 ‹#› .00036 Sometimes the initial centroids will readjust themselves in „right‟ way. then probability = 10!/1010 = 0. – Chance is relatively small when K is large – If clusters are the same size. n. and sometimes they don‟t – Consider an example of five pairs of clusters © Tan.

10 Clusters Example Iteration 1 4 3 2 8 6 4 2 y 0 -2 -4 -6 0 5 10 15 20 x Starting with two initial centroids in one cluster of each pair of clusters © Tan. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Steinbach.

10 Clusters Example Iteration 1 8 6 4 2 8 6 4 2 Iteration 2 y 0 -2 -4 -6 y 0 5 10 15 20 0 -2 -4 -6 0 5 10 15 20 Iteration 3 8 6 4 2 8 6 4 2 x Iteration 4 x y 0 -2 -4 -6 y 0 5 10 15 20 0 -2 -4 -6 0 5 10 15 20 x x Starting with two initial centroids in one cluster of each pair of clusters © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

Kumar Introduction to Data Mining 4/18/2004 ‹#› . while other have only one.10 Clusters Example Iteration 1 4 3 2 8 6 4 2 y 0 -2 -4 -6 0 5 10 15 20 x Starting with some pairs of clusters having three initial centroids. © Tan.Steinbach.

Kumar Introduction to Data Mining 4/18/2004 ‹#› .10 Clusters Example Iteration 1 8 6 4 2 8 6 4 2 Iteration 2 y 0 -2 -4 -6 y 0 5 10 15 20 0 -2 -4 -6 0 8 6 4 2 5 10 15 20 x Iteration 3 8 6 4 2 x Iteration 4 y 0 -2 -4 -6 y 0 5 10 15 20 0 -2 -4 -6 0 5 10 15 20 x x Starting with some pairs of clusters having three initial centroids. © Tan. while other have only one.Steinbach.

Kumar Introduction to Data Mining 4/18/2004 ‹#› .Solutions to Initial Centroids Problem   Multiple runs – Helps. but probability is not on your side Sample and use hierarchical clustering to determine initial centroids  Select more than k initial centroids and then select among these initial centroids – Select most widely separated Postprocessing  Bisecting K-means  – Not as susceptible to initialization issues © Tan.Steinbach.

 © Tan. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Handling Empty Clusters  Basic K-means algorithm can yield empty clusters Several strategies – Choose the point that contributes most to SSE – Choose a point from the cluster with the highest SSE – If there are several empty clusters.Steinbach. the above can be repeated several times.

Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Updating Centers Incrementally  In the basic K-means algorithm. centroids are updated after all points are assigned to a centroid An alternative is to update the centroids after each assignment (incremental approach) – – – – – Each assignment updates zero or two centroids More expensive Introduces an order dependency Never get an empty cluster Can use “weights” to change the impact  © Tan.

Kumar . clusters with relatively high SSE – Merge clusters that are „close‟ and that have relatively low SSE – Can use these steps during the clustering process  ISODATA Introduction to Data Mining 4/18/2004 ‹#› © Tan. i.Steinbach.Pre-processing and Post-processing  Pre-processing – Normalize the data – Eliminate outliers  Post-processing – Eliminate small clusters that may represent outliers – Split „loose‟ clusters..e.

Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Bisecting K-means  Bisecting K-means algorithm – Variant of K-means that can produce a partitional or a hierarchical clustering © Tan.

Bisecting K-means Example © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

Steinbach.Limitations of K-means  K-means has problems when clusters are of differing – Sizes – Densities – Non-globular shapes  K-means has problems when the data contains outliers. © Tan. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

Limitations of K-means: Differing Sizes Original Points K-means (3 Clusters) © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

Kumar Introduction to Data Mining 4/18/2004 ‹#› .Steinbach.Limitations of K-means: Differing Density Original Points K-means (3 Clusters) © Tan.

Kumar Introduction to Data Mining 4/18/2004 ‹#› .Steinbach.Limitations of K-means: Non-globular Shapes Original Points K-means (2 Clusters) © Tan.

but need to put together.Steinbach. © Tan. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Overcoming K-means Limitations Original Points K-means Clusters One solution is to use many clusters. Find parts of clusters.

Kumar Introduction to Data Mining 4/18/2004 ‹#› .Overcoming K-means Limitations Original Points K-means Clusters © Tan.Steinbach.

Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Overcoming K-means Limitations Original Points K-means Clusters © Tan.

Hierarchical Clustering
Produces a set of nested clusters organized as a hierarchical tree  Can be visualized as a dendrogram

– A tree like diagram that records the sequences of merges or splits
6
0.2

5

4 3 2 5 2 1 3 1 4

0.15

0.1

0.05

0

1

3

2

5

4

6

© Tan,Steinbach, Kumar

Introduction to Data Mining

4/18/2004

‹#›

Strengths of Hierarchical Clustering

Do not have to assume any particular number of clusters
– Any desired number of clusters can be obtained by „cutting‟ the dendogram at the proper level

They may correspond to meaningful taxonomies
– Example in biological sciences (e.g., animal kingdom, phylogeny reconstruction, …)

© Tan,Steinbach, Kumar

Introduction to Data Mining

4/18/2004

‹#›

Hierarchical Clustering

Two main types of hierarchical clustering
– Agglomerative:

Start with the points as individual clusters

At each step, merge the closest pair of clusters until only one cluster (or k clusters) left

– Divisive:
 

Start with one, all-inclusive cluster

At each step, split a cluster until each cluster contains a point (or there are k clusters) 

Traditional hierarchical algorithms use a similarity or distance matrix
– Merge or split one cluster at a time
Introduction to Data Mining 4/18/2004 ‹#›

© Tan,Steinbach, Kumar

 Update the proximity matrix Until only a single cluster remains Key operation is the computation of the proximity of two clusters – Different approaches to defining the distance between clusters distinguish the different algorithms Introduction to Data Mining 4/18/2004 ‹#› © Tan.Steinbach. Kumar . 2.Agglomerative Clustering Algorithm   More popular hierarchical clustering technique Basic algorithm is straightforward 1. 4. Compute the proximity matrix Let each data point be a cluster Repeat Merge the two closest clusters 5. 6. 3.

. Proximity Matrix . p1 p2 p3 p4 p9 p10 p11 p12 © Tan. Kumar Introduction to Data Mining 4/18/2004 ‹#› ..Steinbach.Starting Situation  Start with clusters of individual points and a proximity matrix p1 p2 p3 p4 p5 p1 . ... p2 p3 p4 p5 . .

Intermediate Situation  After some merging steps. Kumar Introduction to Data Mining 4/18/2004 ‹#› ..Steinbach.. p1 p2 p3 p4 p9 p10 p11 p12 © Tan. we have some clusters C1 C1 C2 C3 C3 C4 C4 C5 C2 C3 C4 C5 Proximity Matrix C1 C2 C5 .

.Intermediate Situation  We want to merge the two closest clusters (C2 and C5) and update the proximity matrix.. p1 p2 p3 p4 p9 p10 p11 p12 © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› . C1 C2 C3 C4 C5 C1 C2 C3 C3 C4 C4 C5 Proximity Matrix C1 C2 C5 .

Kumar Introduction to Data Mining 4/18/2004 ‹#› .After Merging  The question is “How do we update the proximity matrix?” C1 C1 C3 C4 C2 U C5 C3 C4 C1 ? C2 U C5 ? ? ? ? ? ? C3 C4 Proximity Matrix C2 U C5 . p1 p2 p3 p4 p9 p10 p11 p12 © Tan.Steinbach...

MAX .. Proximity Matrix Distance Between Centroids Other methods driven by an objective function – Ward‟s Method uses squared error p5 © Tan. Group Average . Kumar Introduction to Data Mining 4/18/2004 ‹#› . Similarity? p1 p2 p3 p4      MIN .Steinbach.How to Define Inter-Cluster Similarity p1 p2 p3 p4 p5 ..

     MIN . Kumar Introduction to Data Mining 4/18/2004 ‹#› .. MAX . Proximity Matrix Distance Between Centroids Other methods driven by an objective function – Ward‟s Method uses squared error p5 © Tan. Group Average .How to Define Inter-Cluster Similarity p1 p1 p2 p3 p4 p2 p3 p4 p5 .Steinbach..

     MIN .How to Define Inter-Cluster Similarity p1 p1 p2 p3 p4 p2 p3 p4 p5 .. MAX . Proximity Matrix Distance Between Centroids Other methods driven by an objective function – Ward‟s Method uses squared error p5 © Tan.Steinbach. Group Average .. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

. Proximity Matrix Distance Between Centroids Other methods driven by an objective function – Ward‟s Method uses squared error p5 © Tan..Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› . Group Average .How to Define Inter-Cluster Similarity p1 p1 p2 p3 p4 p2 p3 p4 p5 .      MIN . MAX .

How to Define Inter-Cluster Similarity p1 p1   p2 p3 p4 p2 p3 p4 p5 ... Group Average . Proximity Matrix Distance Between Centroids Other methods driven by an objective function – Ward‟s Method uses squared error p5 © Tan. Kumar Introduction to Data Mining 4/18/2004 ‹#› . MAX .      MIN .Steinbach.

i.80 1.00 I1 I2 I3 I4 I5 1 2 3 4 4/18/2004 5 ‹#› © Tan.e. I1 1.80 I5 0.90 1.60 0.50 0.50 I3 0.90 0.30 0.65 0..Steinbach.65 0.00 0.60 0.20 0.40 1.70 1.10 0.00 0.40 0.20 I2 0.10 0. by one link in the proximity graph.Cluster Similarity: MIN or Single Link  Similarity of two clusters is based on the two most similar (closest) points in the different clusters – Determined by one pair of points.00 0. Kumar Introduction to Data Mining .00 0.30 I4 0.70 0.

Steinbach.Hierarchical Clustering: MIN 5 1 3 5 2 4 4 0 3 6 2 5 4 1 0.15 0. Kumar Introduction to Data Mining Dendrogram 4/18/2004 ‹#› .05 Nested Clusters © Tan.1 0.2 2 3 1 6 0.

Strength of MIN Original Points Two Clusters • Can handle non-elliptical shapes ‹#› © Tan. Kumar Introduction to Data Mining 4/18/2004 .Steinbach.

Kumar Introduction to Data Mining 4/18/2004 .Limitations of MIN Original Points Two Clusters • Sensitive to noise and outliers ‹#› © Tan.Steinbach.

00 0.30 0.Steinbach.80 I5 0.00 © Tan.00 0.30 I4 0.00 0.65 0.40 1.20 0.60 0.40 0.00 0.60 0.90 1.90 0.70 1.20 I2 0.10 0.Cluster Similarity: MAX or Complete Linkage  Similarity of two clusters is based on the two least similar (most distant) points in the different clusters – Determined by all pairs of points in the two clusters I1 I2 I3 I4 I5 I1 1.50 I3 0.65 0.50 0.10 0.80 1.70 0. Kumar Introduction to Data Mining 1 2 3 4 5 ‹#› 4/18/2004 .

3 0.Hierarchical Clustering: MAX 4 2 5 2 3 3 4 1 5 0.2 0.1 0.4 0.15 0.Steinbach.25 6 0. Kumar Introduction to Data Mining 4/18/2004 ‹#› .35 0.05 0 3 6 4 1 2 5 1 Nested Clusters Dendrogram © Tan.

Kumar Introduction to Data Mining 4/18/2004 .Steinbach.Strength of MAX Original Points Two Clusters • Less susceptible to noise and outliers ‹#› © Tan.

Kumar Introduction to Data Mining Two Clusters 4/18/2004 ‹#› .Steinbach.Limitations of MAX Original Points •Tends to break large clusters •Biased towards globular clusters © Tan.

10 0.90 0.00 0.20 I2 0. Clusterj )  piClusteri p jCluster j  proximity(p .70 1.80 1.00 1 2 3 4 4/18/2004 5 ‹#› © Tan. p ) i j |Clusteri ||Clusterj |  Need to use average connectivity for scalability since total proximity favors large clusters I1 I2 I3 I4 I5 I1 1.80 I5 0. proximity(Clusteri .30 0.70 0.60 0.00 0.60 0.50 0.50 I3 0.00 0.65 0.00 0.Steinbach. Kumar Introduction to Data Mining .10 0.30 I4 0.40 0.40 1.90 1.65 0.20 0.Cluster Similarity: Group Average  Proximity of two clusters is the average of pairwise proximity between points in the two clusters.

2 0.1 0.Hierarchical Clustering: Group Average 5 2 5 2 4 1 0.15 3 1 6 0.05 0 4 3 3 6 4 1 2 5 Nested Clusters Dendrogram © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .25 0.

Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Hierarchical Clustering: Group Average  Compromise between Single and Complete Link Strengths – Less susceptible to noise and outliers   Limitations – Biased towards globular clusters © Tan.

– N is the number of points.  O(N3) time in many cases – There are N steps and at each step the size. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Steinbach.Hierarchical Clustering: Time and Space requirements  O(N2) space since it uses the proximity matrix. N2. proximity matrix must be updated and searched – Complexity can be reduced to O(N2 log(N) ) time for some approaches © Tan.

it cannot be undone No objective function is directly minimized Different schemes have problems with one or more of the following: – Sensitivity to noise and outliers – Difficulty handling different sized clusters and convex shapes – Breaking large clusters   © Tan. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Steinbach.Hierarchical Clustering: Problems and Limitations  Once a decision is made to combine two clusters.

MST: Divisive Hierarchical Clustering  Build MST (Minimum Spanning Tree) – Start with a tree that consists of any point – In successive steps. look for the closest pair of points (p.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› . q) such that one point (p) is in the current tree but the other (q) is not – Add q to the tree and put an edge between p and q © Tan.

MST: Divisive Hierarchical Clustering  Use MST for constructing hierarchy of clusters © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

– Density = number of points within a specified radius (Eps) – A point is a core point if it has more than a specified number of points (MinPts) within Eps  These are points that are at the interior of a cluster – A border point has fewer than MinPts within Eps. Kumar Introduction to Data Mining 4/18/2004 ‹#› . but is in the neighborhood of a core point A noise point is any point that is not a core point or a border point.Steinbach. – © Tan.DBSCAN  DBSCAN is a density-based algorithm.

DBSCAN: Core. Border.Steinbach. and Noise Points © Tan. Kumar Introduction to Data Mining 4/18/2004 ‹#› .

Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .DBSCAN Algorithm Eliminate noise points  Perform clustering on the remaining points  © Tan.

Border and Noise Points Original Points Point types: core.DBSCAN: Core. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Steinbach. MinPts = 4 © Tan. border and noise Eps = 10.

When DBSCAN Works Well Original Points Clusters • Resistant to Noise • Can handle clusters of different shapes and sizes © Tan. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Steinbach.

Steinbach.DBSCAN: Determining EPS and MinPts    Idea is that for points in a cluster. their kth nearest neighbors are at roughly the same distance Noise points have the kth nearest neighbor at farther distance So. Kumar Introduction to Data Mining 4/18/2004 ‹#› . plot sorted distance of every point to its kth nearest neighbor © Tan.

Steinbach. Kumar . recall  For cluster analysis. precision.Cluster Validity  For supervised classification we have a variety of measures to evaluate how good our model is – Accuracy. the analogous question is how to evaluate the “goodness” of the resulting clusters? But “clusters are in the eye of the beholder”! Then why do we want to evaluate them? – – – – To avoid finding patterns in noise To compare clustering algorithms To compare two sets of clusters To compare two clusters Introduction to Data Mining 4/18/2004 ‹#›   © Tan.

5 0.7 1 0.4 0.5 0.3 0.8 1 x x © Tan.6 0.6 0.6 0.2 0.6 0.6 0.4 0.5 0.4 0.4 0.9 0.8 1 Random Points 0.8 0.7 0.2 0.7 0.8 0.4 0.3 0.8 0.9 0.4 0.2 0.2 0.9 0.8 1 0 0 0.3 0.2 0.4 x K-means 0.Clusters found in Random Data 1 0.8 1 x 1 0.5 0.Steinbach.6 0.8 0.6 0.4 0.2 0.1 0 0.6 0.1 0 0.1 0 Complete Link y y 0.9 1 0.2 0. Kumar Introduction to Data Mining 4/18/2004 ‹#› .7 0.2 0.3 0.1 0 DBSCAN y y 0 0 0.

3. Comparing the results of a cluster analysis to externally known results.. Determining the clustering tendency of a set of data. 3. © Tan. distinguishing whether non-random structure actually exists in the data. 5. Determining the „correct‟ number of clusters.e. i.Steinbach. e. to externally given class labels. . Kumar Introduction to Data Mining 4/18/2004 ‹#› .g. For 2. we can further distinguish whether we want to evaluate the entire clustering or just individual clusters. Comparing the results of two different sets of cluster analyses to determine which is better. and 4.Different Aspects of Cluster Validation 1.Use only the data 2. Evaluating how well the results of a cluster analysis fit the data without reference to external information. 4..

Kumar Introduction to Data Mining 4/18/2004 ‹#› .Measures of Cluster Validity  Numerical measures that are applied to judge various aspects of cluster validity. – External Index: Used to measure the extent to which cluster labels match externally supplied class labels.g.  Sum of Squared Error (SSE) – Relative Index: Used to compare two different clusterings or clusters.. SSE or entropy  Sometimes these are referred to as criteria instead of indices – However.  Often an external or internal index is used for this function.  Entropy – Internal Index: Used to measure the goodness of a clustering structure without respect to external information. sometimes criterion is the general strategy and index is the numerical measure that implements the criterion. e. © Tan. are classified into the following three types.Steinbach.

Kumar .   High correlation indicates that points that belong to the same cluster are close to each other. only the correlation between n(n-1) / 2 entries needs to be calculated.Steinbach.Measuring Cluster Validity Via Correlation  Two matrices – – Proximity Matrix “Incidence” Matrix    One row and one column for each data point An entry is 1 if the associated pair of points belong to the same cluster An entry is 0 if the associated pair of points belongs to different clusters  Compute the correlation between the two matrices – Since the matrices are symmetric. Introduction to Data Mining 4/18/2004 ‹#› © Tan. Not a good measure for some density or contiguity based clusters.

5 0.4 0.8 0.4 0.9 0. 1 0.6 0.2 0.6 0.6 0.9 0.2 0.3 0.8 1 y x y x Corr = 0.Steinbach.4 0.5810 4/18/2004 ‹#› .5 0.3 0.4 0.Measuring Cluster Validity Via Correlation  Correlation of incidence and proximity matrices for the K-means clusterings of the following two data sets.6 0.7 0.8 0.1 0 0 0.8 1 1 0.2 0.2 0.7 0.1 0 0 0.9235 © Tan. Kumar Introduction to Data Mining Corr = 0.

Steinbach.2 0.1 20 40 60 80 0 100 Similarity Points 0 0.8 0.9 0.1 0 y 50 60 70 80 90 100 x Points © Tan. 1 0.2 0.8 1 0.7 0.9 0.4 0.6 0.6 0.7 10 20 30 40 1 0.Using Similarity Matrix for Cluster Validation  Order the similarity matrix with respect to cluster labels and inspect visually.8 0.3 0. Kumar Introduction to Data Mining 4/18/2004 ‹#› .3 0.5 0.5 0.4 0.2 0.4 0.6 0.

1 20 40 60 80 0 100 Similarity 1 0.9 0.7 0.2 0. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Using Similarity Matrix for Cluster Validation  Clusters in random data are not so crisp 1 10 20 30 40 0.4 0.3 0.2 0.8 0.7 0.4 0.1 0 0 0.6 0.3 0.Steinbach.6 0.2 0.6 0.5 0.8 0.9 0.4 0.5 0.8 1 Points 50 60 70 80 90 100 Points y x DBSCAN © Tan.

1 20 40 60 80 0 100 Similarity 1 0.9 0.4 0.4 0.9 0.2 0.5 0.6 0.2 0.6 0.8 0.8 0.2 0.5 0.Using Similarity Matrix for Cluster Validation  Clusters in random data are not so crisp 1 10 20 30 40 0.6 0.4 0.3 0.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .8 1 Points 50 60 70 80 90 100 y Points x K-means © Tan.7 0.7 0.3 0.1 0 0 0.

4 0.5 0.4 0.7 0.5 0.9 0.9 0.3 0.4 0.2 0.8 0.6 0. Kumar Introduction to Data Mining 4/18/2004 ‹#› .2 0.1 20 40 60 80 0 100 Similarity 1 0.7 0.6 0.2 0.3 0.1 0 0 0.6 0.Using Similarity Matrix for Cluster Validation  Clusters in random data are not so crisp 1 10 20 30 40 0.8 1 Points 50 60 70 80 90 100 Points y x Complete Link © Tan.Steinbach.8 0.

1 5 2500 7 3000 500 1000 1500 2000 2500 3000 0 DBSCAN © Tan.7 0. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Using Similarity Matrix for Cluster Validation 1 0.4 2000 0.8 0.Steinbach.2 0.9 1 2 6 500 0.5 0.3 0.6 1000 3 4 1500 0.

Can also be used to estimate the number of clusters 10 6 4 2 9 8 7 6 SSE 5 4 0 -2 -4 -6 5 10 15 3 2 1 0 2 5 10 15 20 25 30 K © Tan.Steinbach. Kumar Introduction to Data Mining 4/18/2004 ‹#› .Internal Measures: SSE   Clusters in more complicated figures aren‟t well separated Internal Index: Used to measure the goodness of a clustering structure without respect to external information – SSE   SSE is good for comparing two clusterings or two clusters (average SSE).

Kumar Introduction to Data Mining 4/18/2004 ‹#› .Steinbach.Internal Measures: SSE  SSE curve for a more complicated data set 1 2 6 3 4 5 7 SSE of clusters found using K-means © Tan.

Internal Measures: Cohesion and Separation  Cluster Cohesion: Measures how closely related are objects in a cluster – Example: SSE   Cluster Separation: Measure how distinct or wellseparated a cluster is from other clusters Example: Squared Error – Cohesion is measured by the within cluster sum of squares (SSE) WSS    ( x  mi )2 i xC i – Separation is measured by the between cluster sum of squares BSS   Ci ( m  mi ) i © Tan. Kumar 2 – Where |Ci| is the size of cluster i Introduction to Data Mining 4/18/2004 ‹#› .Steinbach.

Internal Measures: Cohesion and Separation

Example: SSE
– BSS + WSS = constant
m 3 4


1 m1 2 K=1 cluster:


m2 5

WSS (1  3)2  (2  3)2  (4  3)2  (5  3)2  10 BSS 4  (3  3)2  0 Total  10  0  10

K=2 clusters:

WSS (1  1.5)2  (2  1.5)2  (4  4.5)2  (5  4.5)2  1 BSS 2  (3  1.5)2  2  (4.5  3)2  9 Total  1  9  10

© Tan,Steinbach, Kumar

Introduction to Data Mining

4/18/2004

‹#›

Internal Measures: Cohesion and Separation

A proximity graph based approach can also be used for cohesion and separation.
– Cluster cohesion is the sum of the weight of all links within a cluster. – Cluster separation is the sum of the weights between nodes in the cluster and nodes outside the cluster.

cohesion

separation
‹#›

© Tan,Steinbach, Kumar

Introduction to Data Mining

4/18/2004

Internal Measures: Silhouette Coefficient
 

Silhouette Coefficient combine ideas of both cohesion and separation, but for individual points, as well as clusters and clusterings For an individual point, i
– Calculate a = average distance of i to the points in its cluster – Calculate b = min (average distance of i to points in another cluster) – The silhouette coefficient for a point is then given by s = 1 – a/b if a < b,
(or s = b/a - 1 if a  b, not the usual case)
b a

– Typically between 0 and 1. – The closer to 1 the better.

Can calculate the Average Silhouette width for a cluster or a clustering
Introduction to Data Mining 4/18/2004 ‹#›

© Tan,Steinbach, Kumar

Kumar Introduction to Data Mining 4/18/2004 ‹#› .External Measures of Cluster Validity: Entropy and Purity © Tan.Steinbach.

Steinbach.” Algorithms for Clustering Data.Final Comment on Cluster Validity “The validation of clustering structures is the most difficult and frustrating part of cluster analysis. Without a strong effort in this direction. Kumar Introduction to Data Mining 4/18/2004 ‹#› . Jain and Dubes © Tan. cluster analysis will remain a black art accessible only to those true believers who have experience and great courage.

33 .943           90 7. :2.7 3974/:.79399439.943                  90 7.3 $903-.:89078 %..741.:8907410.94394.03974/83430.943          90 7.9.9:.473 90 7..3:.943                     $9..5.

.

  .

9..943                  $9.3970039.03974/8 04907.:89078.7841.3:.3 $903-..9:.043430 %.33 .94394.7 3974/:.. :2..473 90 7.793984205.

.

  .

943                           90 7..:89078.3 $903-.7841.473 90 7.7 3974/:.3:.943   90 7.793984205.94394.03974/8 04907.043430 %.9:.9..943                    $9.. :2.943          90 7.3970039.33 ..

.

  .

90/ !489574..35.3/:8007.0883  80.-98349434:78/0 $.243908039...65:..94388:08 %.250.7 3974/:.3 $903-.&63.93 20..059-09439..03974/8  $00..33 .9.94394.3/903 800. :2.03974/8  $00.9.339.88:8.:8907394 /090723039...38  49.65.7.961:#96/34  :9507:38 058 -:9574-.7.924709.92489/0805.03974/8.

.

  .

7080.9008 44809054399.9..3:..513547.3 $903-.4792.07.-4.33 ..:8907990089$$ 19070. :2.0.:89078 90.8.38.9208  %.:89078 $0.4397-:908248994$$ 4480.07.07.9:  .0259.7 3974/:.897.30/0259 .9.3-0 7050.94394.. 20.90/80.5439174290.

.

  .

8832039 3.8830/94.5574..7 3974/:.:8907 .9080744794.30259.0 3974/:. .9  %.9.08.70...03974/8.9073...90/.3:80 098 94.0894:5/.9090. 20.1907.03974/ 3.309025.8..4792 .33  390-....70 :5/.03974/8 47005038.94394..347/07/0503/03. :2.54398.33 .3 $903-.8832039:5/.0709. 0..9.03974/8.55.9:50945.702039..38.71.1907 0.

.

  .

#9.

51#6:.7960::5.

070.480 .7960::5  !70 574..3/9.9.:89078  0 .0883 23.9.33 .:89078970.088 $ % 3974/:.9.7057080394:9078 $59 4480 .904:9078  !489 574..0 $$ 070.70 .:890789.9.92.3:80908089058/:7390.9.9082.0 4$$ . 23.:89073574.9.0883 472.:890789.090/.94394.

.

7 . :2.  %.3 $903-..

5.:0.

5.7. 07.33 .47.0.3574/:..7 3974/:.7.3 $903-.9.:89073 %.9.38.389.5:  80.93 20.4.79943..94394. :2..3941 20..4792 '..

.

  .

:0.5.

.4.473 %. :2.7 3974/:.33 .5:.9.94394.3 $903-.

.

  .

4.65:6...

38 4:9078 %.7 3974/:.94394..38.8574-0280390/.8574-02803. :2.7041 /11073 $08 038908 43 4-:.439.508   20.9.38.9.78..5:   20.4.:89078.33 .3 $903-.

.

  .

.4.65:6..

38 :89078 %.5:95&: 73.3 $903-.!4398  20.4.94394.7 3974/:. :2..33 .9.

.

  .

4...65:6.

 :2.!4398  20.7 3974/:.33 . 73.9.5:955:..38 :89078 %.94394.4.3 $903-.

.

  .

4..65:6..

5:!65.4.

9&.38 :89078 %.7: 73.. :2.!4398  20.7 3974/:.36/3.94394.3 $903-.33 .9.

.

  .

"90645.

9.4.5:4...38:89078 3084:943894:802.:89078 -:9300/945:9940907 %..!4398  20.3.7 3974/:.33 . :2.3 $903-.65: 73.:89078 3/5.94394.79841.

.

  .

"90645.

.94394.65: 73.!4398  20.5:4.33 . :2.9.38:89078 %.4.3 $903-...7 3974/:.

.

  .

"90645.

9.38:89078 %. :2.94394..5:4.4...33 .65: 73.7 3974/:.3 $903-.!4398  20.

.

  .

94394.8.8.29.3-0.7.33 . :2.970.900.7 3974/:.08..9.7.30/.33:.0841 20708478598                       %.47/890806:03.2  97000/.9.9700  . 07.0/.95 !74/:.:8907847./03/747..3 $903-..8094130890/.8:.

.

  .

434208 ..33:.73:2-0741 .. :2.3/42  540370.0  %02..33 .094.47708543/9420.3 $903-..8.2503-44.:89078 3/0870/3:2-0741.3-04-9.9905745070.95  4349.32.88:20.08 0  ..95.30/- .:993 90/03/47.43897:.:.&.94394.:69.943  %.9.:89078.35.79.2.900.7 3974/:.331:9.03.

.

  .

/943.8905 207090.395084107.82.33:.0 $9.439.:8907.7947 /89.741.:8907:390..70.:89078  %7..38.5439 47 9070.9.:8907 47.799430 ...:89078 019 .83/.3.97 07047859430.900.33 .02..07.:89073 4207....47928:80.9.:8907.:8.:89078 90.8.0.95  %42.:89078:3943430.480895.920 3974/:.7.8905 859.../:..7.:8907 90.7999054398.94394..9. 3.9.0 $9.

.

  %.7 .. :2.3 $903-.

. :2.97 090.7 3974/:..9090574292.47928 %.48089.:89078 &5/.7.0-09003 .3.38  04507..425:9....94394.47928897.36:0 ...33 .:89078 1107039.94341905742941 94.95369.3:.5574.8.4   470545:.3649.97 &39 43.707.9 0709094.:8907 #050.9.9147.:89078/893:890/1107039.5439-0.0894/013390/89.:8907702.7/       425:9090574292.830.943890.9.:8907390./..3 $903-.

.

  .

97 5 5 5 5 5 5 5 5 5 5   !7429.3 $903-...97 %.54398.94394.5&.:89078413/.7 3974/:. :2.65  $9..3/.33 .9.9..&.799. 574292./:.

.

  .

941.:89078             !7429. :2.3 $903-.33 .5.&.9.94394....7 3974/:..97    %..65  19078420207389058 0.08420.

.

  .

3/ :5/.3 $903-.48089. :2.5.3/ ..65  0.97    %.94394.9090574292.33 .:89078 .9.7 3974/:...97              !7429..399420709094.941.&.

.

  .

 :2.9.9 95  %06:089438 4/40:5/.94394...33 .9090574292.97 &  %.7 3974/:.97    &         &      !7429.3 $903-.

.

  .

6.655.9.

7/ 8094/:80886:.9&43.9.9.07.3.79 5 5 5 5        74:5.00900303974/8 9072094/8/7.7 3974/:. :2. 5 5 5 5 5   $2.3:..97 89.94394.03-.70/07747 5 %.0 !7429.34-0.33 .0 1:3.9.3 $903-.943 .

.

  .

6.655.9.

03-.7 3974/:.9&43.07.3:.33 .3 $903-. 5 5 5 5 5 5 5 5 5          74:5.97 89..0 !7429.70/07747 5 %.3.0 1:3.943 .34-0.9.94394.9. :2.7/ 8094/:80886:.9.00900303974/8 9072094/8/7.

.

  .

9.6.655.

. 5 5 5 5 5 5 5 5 5          74:5.34-0.3:.94394.9.33 .9.9&43.0 !7429.7/ 8094/:80886:.97 89.943 .7 3974/:.07.03-.3.00900303974/8 9072094/8/7.70/07747 5 %.0 1:3.9. :2.3 $903-.

.

  .

655.9.6.

.3:.00900303974/8 9072094/8/7.03-.943 .7/ 8094/:80886:.7 3974/:.9.33 .3 $903-.0 !7429. :2.9.97 89.34-0.9.0 1:3.70/07747 5 %. 5 5 5 5 5 5 5 5 5          74:5.94394.9&43.3.07.

.

  .

9.655.6.

03-. :2.0 !7429.0 1:3.9.70/07747 5 %..7 3974/:.3:.3.9.9.9&43.34-0.943 .07.33 . 5 5 L L 5 5 5 5 5 5 5          74:5.97 89.3 $903-.7/ 8094/:80886:.94394.00900303974/8 9072094/8/7.

.

  .

5                                         %.9.:89078 0907230/-4305.794194.3 $903-.3:.9&43.:890788-..7 3974/:. !69&5352  $2.7 .9.33 .48089 54398390/1107039 .94394.80/439094 248982. :2.74154398  0 -43033 90574297.

.

  .

9.3 $903-. :2..9.33 03/747.95 !                       0890/:89078 %.33:.2 .900.94394.7 3974/:.

.

  .

95..&.6 ! 73..3/0343 059.3 $903-. :2.94394.9.7 3974/:.508 %.!4398 %4:89078 W .33 .8.3.

.

  .

4...9.7 3974/:.0943480.65:6 ! 73.33 .3 $903-.3/4:9078 %.. :2.!4398 %4:89078 W $0389.94394.

.

  .

94394.:89078            %..7                              3974/:.:890788-.52.9.7 2489/89.78415439839094.9&43.794194.39 54398390/1107039 .33 .89 82.9.80/4390940. 696473. :2.  $2.:89078 0907230/-.3:.5.3 $903-.

.

  .

.9.900. :2.7 3974/:.33 .33:.9.94394.95                              0890/:89078 03/747.2 %.3 $903-.

.

  .

&..95.!4398 %4:89078 W 0888:8.3/4:9078 %.33 .7 3974/:.9.94394. :2.059-0943480.3 $903-.6  73.

.

  .

70.!4398 %4:89078 W%03/894-70.7....4.:89078 W.80/94.:89078 %.94394.7 3974/:.9.7/84-:.65:6  73. :2..3 $903-.33 .

.

  .

  !74294194.07..9.0949.3:.9.9.9  7Z3:.9 3:.0.94394.7 3974/:..:89078                                         %. :2.0415. 3:. 7 7  7964.478.3 $903-.:89078890.9 =3:...78057429 -090035439839094.33 .70.-983. 574291.9&43.4330.:89078 7964.07.9 =  00/94:80.91478.9679.9 7 Z3:.9 = =3:...

.

  .

 :2.33 .94394.                         0890/:89078 03/747.9.959679.9.3 $903-..900.33:.7 3974/:.2 %.

.

  .

33:.7/84-:.  42574280-09003$30.9.7.:89078 %.9.7 3974/:.900.94394.33 .3/4:9078   29.9438 .3/425090 3 $970398 0888:8.80/94..059-0943480. :2.3 $903-.959679.

.

  .

08 %.5574.33:.90/..3.3-070/:.083.808 %070.3/80..7089058..0/94  4   920147 8420.:   85.9.51&7..0/ 42509..098945.9.3 $903-.95'4.94394.7.09:80890574292.33 .90.3/. :2..97  8903:2-074154398   92032.972:89-0:5/.89059080   574292.7 3974/:.900.

.

  .

3.3/3/110703980/.9.92320/ 11070398./094.3/.9.84382.9...:9.01:3.9438/70. :2.:89078  9..0 8.:89078   %.:89078.3349-0:3/430 44-0.3 $903-.33:./0.94394.0.42-3094.0574-028943047 247041901443 $0389.514.70.9943480..0208.3/4:9078 11.508 70.33 .65:  3.43..7 3974/:.900.95#96/34:.

.

  .

333%700 $9.088..9.30/0-090035.94305439 5 8390.089058 4414790.3 $903-.&':9. 9.9.35439 38:.480895. :2.94394.7 3974/:.900.43889841..:770399700-:9904907 6 8349 //694909700.95  :/$% 32:2$5.3/5:9.74154398 5 6 8:.33 .33:.799.97009.3/6 %.

.

  .

9307.&':9. :2.3 $903-.7.94394.33:.33 .900.:89078 %.43897:..7 3974/:.95  &80$%147.9.41.

.

  .

4705439 34805439 8.94394.10/7.470543947./0389 -.:8907 -47/075439 .810079./:8 58 54398.3.354399.3 $903-.&!  $8.4792 03893:2-07415439893.9.824709.. :2.33!989358 -:983 9030-4744/41..850.33 .9.-47/07 5439  %.99039074741.70543989.850.80/.98349.10/3:2-07 4154398 3!98 9358 %080.70.7 3974/:..4705439 19...

.

  .

&!69 6919 .7 3974/:.: %. :2.51!6:#65.3 $903-.9..94394.33 .

.

  .

3354398  %.&!369.9. :2..4 23.94394.90348054398  !071472.7 3974/:.:890734390702.33 .3 $903-.

.

  .

94394.3 $903-. :2.51!6:#65.&!69 6919..: 73.470  -47/07 .7 3974/:.9.3/3480 58 3!98 %.!4398 !4399508.33 .

.

  .

94394.508.3.5&!692:33 73.:8907841/11070398.!4398 :89078 W #0889..33 . :2.3994480 W .9.3/808 %.3 $903-.7 3974/:.3/0.

.

  .

 :2.9147543983..20/89.70.:    /0.0 $4 54984790//89.0410.3.&!.:8907 9079 30..51 5#.9455#&.0909 30.974:908.94394.708930-47 %.3.89.3..7 3974/:.7089 30-478.3 $903-.708930-47.07543994989 30.7907 /89.33 .0 48054398.91.9.

.

  .

425.47928 %4..33 .3:.:..9090 44/3088 4190708:93.94394.. :2.425.:8907..9.843 70.39940.:89078   %.881. 570..  478:507...70941 20.9430.7094.:89078.3 $903-.8:708940..4/13/35.70..31.0.80/.7 3974/:.99073833480 %4.425..3..:.:89078 %4.70390004190-04/07  %03/40.:7.9.44:86:089438494 0.88 90.:.90902 %4.  47.3..7094809841.:89073..90444/4:724/08 .:89078 :9 .

.

  .

3/42 !4398       $                         20.94394.9..38         425090 3        %..3:.5164.                         #.7 3974/:..3 $903-.9:6515%.33 . :2.

.

  .

.:.:89073903/03.-08 .3/ 0.95.9. 41.:.9..80941/.:8907.881990/... 94:9 7010703.943  &804390/.9.7 3974/:.:8907.80894 /0907230.65  090723390 .7390708:984194/1107039809841.:8907347:893/.3.31472.9:.3.9.3 $903-.:89078  %.3/42897:.:8907.9:70.  425.9.     425..03.4770.31. :2.  0  /893:830907343 7.9 3:2-0741.:89078 47  .94394..889409073.343 708:98 0  9409073.909003970./:.31:7907/893:809070...09409073..9.7390708:9841.88..3994 0..0898390 /.33 .:70.8-09907 090723390 .934090708:9841.:63:.3.

.

  .

70.08 40.550/94:/0.850. :2.7907.:8907.92502039890.9.:8907.9.8:550/.99409073.7907438900307.8:7089.98 41.4739073..:89078  1903.90.3/0 &80/9420.707010770/94.-08 2..:89073 897:.9.20.309073.8.943 0  $$47 039745  $42092089080..:89073847 .33 .70.31472.8810/39490144397009508 9073..8:709.9:7094:9 70850..790743 %.88.9.7094/1107039.09073.3/0 &80/9420.9.31..897..943  $:241$6:.03/0 &80/94./9 .8:70900903994.3 $903-.3/08:80/147981:3..94394.8:709044/308841.-08 39745 39073.425.:9:63:.74:8.3/3/0890 3:207.70/7747 $$ #0.07 84209208.20..  :207. 3890../413/.7 3974/:.

.

  .

90/5.08 !7429.08.5439 303978190.:953:.90/5..943-09003 3 3  .74154398-043894/1107039.6993.97 3.3/430.31.08 $3.4770.0 .74154398-04394908.97 3074.4:231470.97.9..70822097./..4770.97.884.9.97../03..884..:89078  425:9090. 4390.:8907 303978 190.65  %42.0902.20.943-0900390942.

.8:701478420/038947.9..9-0439490 8..439:9-.70.:..80/ .94394.480940.33 .9433/.:8907.20.4770.:89078 3974/:.90/   .4907  49.039708300/894-0.44/20.9543989.9089.

.

 :2..  %.3 $903-.7 .

.33 .:953:..9.8098                                    477  %.3/574292./03.31.97.:8907384190144394 /.65  4770..3 $903-.6993.38..943413.0.9.08 14790 20.9.7 477  3974/:. :2.94394.

.

  .

:8907 .8:.31.7 3974/:..9.994. :2.65  7/079082.                         $  2.9693:..9. ..792.9. 79 !4 398                    ! 43 98 %.-08.:5&43.3 $903-.97970850.3/3850.94394.9.33 .

.

  .

65  :8907837...9.:5&43.7034984.3/42/.9.9.9693:.785                    $ 2.31.7 3974/:.9. 79                !4 39 8       ! 43 98   $ %...94394.33 .3 $903-. . :2.

.

  .

 :2.:5&43.94394. .38 %...9.785                    $ 2.7034984.3/42/.33 .9.9693:..9.. 79                !4 39 8        ! 43 98   20.7 3974/:.65  :8907837.3 $903-.9.31.

.

  .

9..7 3974/:.9693:..7034984. 79                !4 39 8       ! 43 98   4250903 %. :2.3/42/.33 .:5&43..94394.9.9.65  :8907837.3 $903-.31.785                    $ 2.. .9.

.

  .

9693:.31.94394..33 .3 $903-. :2. .:5&43.7 3974/:..65                                $ %..9.9.9.

.

  .

31472..90903:2-0741.90/ 39073.3 .07.7.:89078         $$                     %.3/0&80/9420....33 .94394.943 $$   $$844/147.7394.425.:9:&&   :8907832470.8:709044/308841. :2.99409073.9.7 3974/:.90/1:708.95.:89078 .84-0:80/940892.:8907384794.703 90805.:89073 897:.0$$ .3.3 $903-.425.9:7094:970850.5.

.

  .

9.33 .5.9.95.:8907814:3/:83 20.3 $903-. :2.809        $$41.:9:&&  $$..2470.94394.:7..0147.425.90//.3 .38 %.7 3974/:.

.

  .

95.:8907817424907..48070.8:704/893.51&7.943820.7..704-0.7  2 2  0708908041.:89078:24186:..94394.:9:6:65.8:7084.9470 805.:89078 ..5.7.9.90/ .3 $903-.:89078:24186:.7.8:70/-9093.:8907 3974/:.65  :890740843 0.983.33 .708 $$  $$   Z   2  $05.9. :2.3 .:8907 .8:70/-90-09003.90/.250$6:.250$$  :8907$05.9430.708 $$   %.70/7747 40843820.

.

  .

51&7.:9:6:65.3 .      %.3 $903-..7  L       .:8907 $$            $$  L     %49.. :2.:89078 $$                    $$  L    %49.9.250$$ $$ $$.4389.65  .     .95.39 2  L 2  L     2 L   .5.

.

94394.9.  3974/:.33 .

3..3/34/084:98/090.7.84-0:80/147 .5.40843.5574...95.3/805.51&7.:9:6:65.7.3 .:8907 .80/..:8907 :8907805.3 $903-.9.9438908:24190098-0900334/08390.943 :8907.943 %.:8907 .7.65  574297.5-.7 3974/:.33 .3893.9.94394..408438908:241900941..40843 805. :2.

.

  .

039147.:.605.33/.:..041 94543983.80.03- 8 .943  -:91473/.54398903./:.5439  .:9:&36.5.8.:890738 47.42-30/0.3/.   $4:09904011.:89078.0/89.3.34907.:8907 %084:0990./:. .041 949054398398.40843.54398 .3 .07...3/805.90.23 .:8907 .0/89.07.039.95.7...841-49.4011.3.90.

-1. 478-..

. 1.3.. .  .80 %5.48079490-09907 .K .-09003 ...0$4:0990/9147.9.34990:8:.07.3/  %0...33 .:.:890747.94394..9090.:89073 3974/:.

.

.  %.3 $903-.7 . :2.

3 $903-.31.5.51#9.9..94394.33 .95..967.9.7 3974/:.3 . :2.:9:63:. %.

.

  .

94341..088-04394 948097:0-00..79.0.9.3 $903-.88702.:89073897:..3/:-08 %..3.3.33 .9:7088902489 /11.-.:8907 .31.89743011479398/70..3..4:7.:9. %0.7 3974/:.:8907.935.653:./.5. .88  94:9.0784.9.36445..3/17:897.3/ 70.0050703.94394.3.943 .7941. :2.9.0 47928147:89073.9.

.

  .