You are on page 1of 7

Using ML In Regression

Testing
The Plan

• SUT: System Under Testing


• ANA: Analysis scripts
• TD: Testing Data
• LD: Log Data
• AUTO: Autoencoder
• ANN: Artificial Neural
Network
The Plan

• Gather testing and log data on passed regression tests for training.
• Transform these datasets into a format that the neural network can handle.
• Train an artificial neural network to reproduce the log feed of a well-functioning
system.
• Use this ANN as an oracle to create the expected log.
• With distance calculation between the SUT log and the expected log, analysis is
only needed when an anomaly occurs.
• Generated test cases can be used.
Autoencoder

• Feed test cases / test data


into an autoencoder.
• Encoder part can be used to
cluster data into categories
based on similarity.
• Decoder part can be used to
generate new ones based
on these categories.
Clustering

• Clustering algorithms like K-


means can identify data types
based on the encoded data.
• These types can be known
similarities for example
handovers etc.
• Based on the intervals in
these clusters we can
generate new testcases by
feeding random numbers in
the decoder part.
Aftermath

• SUT: System Under Testing


• ANN: Artificial Neural
Network
• AUTO: Autoencoder
• TD: Testing Data
• LD: Log Data
• ED: Expected Data
• DC: Distance Calculation
• ANA: Analysis
Thank you!

You might also like