Adding support for CAND #258
Conversation
|
Failing tests are due to moa.classifiers.meta.imbalanced tests failing (issue #260). |
abifet
left a comment
There was a problem hiding this comment.
Can we have the two classifiers in classifiers/functions like in https://github.com/Waikato/wekaDeeplearning4j/tree/master/src/main/java/weka/classifiers/functions ?
Sure! We already have a separate classifier for MLP (moa/src/main/java/moa/classifiers/neuralNetworks/MLP.java) Were you thinking of another one for RNNs? |
As per the offline conversation, renamed the folder 'neuralNetworks' to 'deeplearning' in commit 4457873. |
…Engine random seed (#263) * Initial version of MLP support using DJL * Initial version of CAND * Unit tests for CAND and MLP * Add reference and description. * Rename the package directory to 'deeplearning' * As per the paper, use the best performing defaults. * Fixing issue #260 - Add items to ADWIN ensemble only if drift detection is enabled. Drift detection is off by default. * Issue #260 -Add new reference files * Add an option to set DJL Engine random seed * Add reference files for MLP and CAND missed in #258
This is a preliminary PR for initial review
TODO
T E S T S
Running moa.integration.SimpleClusterTest
Tests run: 5, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 19.17 sec - in moa.integration.SimpleClusterTest
Running moa.streams.filters.SelectAttributesFilterTest
Tests run: 1, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.038 sec - in moa.streams.filters.SelectAttributesFilterTest
Running moa.classifiers.trees.EFDTTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 2.279 sec - in moa.classifiers.trees.EFDTTest
Running moa.classifiers.trees.DecisionStumpTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.419 sec - in moa.classifiers.trees.DecisionStumpTest
Running moa.classifiers.trees.FIMTDDTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.066 sec - in moa.classifiers.trees.FIMTDDTest
Running moa.classifiers.trees.HoeffdingAdaptiveTreeTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.967 sec - in moa.classifiers.trees.HoeffdingAdaptiveTreeTest
Running moa.classifiers.trees.RandomHoeffdingTreeTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.404 sec - in moa.classifiers.trees.RandomHoeffdingTreeTest
Running moa.classifiers.trees.HoeffdingOptionTreeTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.867 sec - in moa.classifiers.trees.HoeffdingOptionTreeTest
Running moa.classifiers.trees.ORTOTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.052 sec - in moa.classifiers.trees.ORTOTest
Running moa.classifiers.trees.ASHoeffdingTreeTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.673 sec - in moa.classifiers.trees.ASHoeffdingTreeTest
Running moa.classifiers.trees.LimAttHoeffdingTreeTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.399 sec - in moa.classifiers.trees.LimAttHoeffdingTreeTest
Running moa.classifiers.trees.AdaHoeffdingOptionTreeTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 0.787 sec - in moa.classifiers.trees.AdaHoeffdingOptionTreeTest
Running moa.classifiers.trees.HoeffdingTreeTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 1.161 sec - in moa.classifiers.trees.HoeffdingTreeTest
Running moa.classifiers.neuralNetworks.CANDTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 56.387 sec - in moa.classifiers.neuralNetworks.CANDTest
Running moa.classifiers.neuralNetworks.MLPTest
Tests run: 2, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 34.205 sec - in moa.classifiers.neuralNetworks.MLPTest
Running moa.classifiers.meta.imbalanced.OnlineAdaC2Test
Tests run: 2, Failures: 0, Errors: 1, Skipped: 0, Time elapsed: 0.343 sec <<< FAILURE! - in moa.classifiers.meta.imbalanced.OnlineAdaC2Test
testRegression(moa.classifiers.meta.imbalanced.OnlineAdaC2Test) Time elapsed: 0.322 sec <<< ERROR!
java.lang.NullPointerException: Cannot invoke "java.util.ArrayList.add(Object)" because "this.adwinEnsemble" is null
`
CAND and MLP tests are passing. They take a bit more time than the previous ones. But some of the later tests are failing. I doubt it is due to CAND and MLP tests.
Testing
EvaluatePrequential -l (neuralNetworks.CAND -h -n) -s (ArffFileStream -f elecNormNew.arff)EvaluatePrequential -l (neuralNetworks.CAND -h -n) -s (ArffFileStream -f elecNormNew.arff) -f 1000 -q 1000