Forwarding to the @dev list.
--
Gianmarco
---------- Forwarded message ----------
From: Ercan Öztürk <[email protected]>
Date: 7 September 2015 at 16:57
Subject: HoeffdingTree and VerticalHoeffdingTree Classifiers run on KDD Cup
99 Data Set
To: [email protected]
Hi Mr. Morales and Mr. Bifet,
We are a couple of undergrad students from TOBB University. As a data
mining class project, we decided to run HoeffdingTree classifier-in moa and
VerticalHoeffdingTree classifier-in samoa on KDD Cup 99 data set (couldn't
attach the data set to this mail due to the size limitations of the Apache
mail server) and present the results in our project report.
We were able to run HoeffdingTree Algorithm on the KDD Cup 99 (both on
kddcup_full.arff, kddcup_10_percent.arff) data set. VerticalHoeffdingTree
classifier also works fine on kddcup_10_percent.arff. However, when we try
to run the VerticalHoeffdingTree classifier on kddcup_full.arff, we got the
following error:
The command we use to run SAMOA Local:
bin/samoa local target/SAMOA-Local-0.3.0-SNAPSHOT.jar
"PrequentialEvaluation -i -1 -f 41920 -l
(com.yahoo.labs.samoa.learners.classifiers.trees.VerticalHoeffdingTree -p
4) -s (com.yahoo.labs.samoa.moa.streams.ArffFileStream -f kddcup_full.arff)"
The console output of samoa:
bin/samoa
Deploying to LOCAL
Command line string = PrequentialEvaluation -i -1 -f 41920 -l
(com.yahoo.labs.samoa.learners.classifiers.trees.VerticalHoeffdingTree -p
4) -s (com.yahoo.labs.samoa.moa.streams.ArffFileStream -f kddcup_full.arff)
2015-09-01 22:22:16,160 [main] INFO com.yahoo.labs.samoa.LocalDoTask
(LocalDoTask.java:79) - Successfully instantiating
com.yahoo.labs.samoa.tasks.PrequentialEvaluation
2015-09-01 22:22:17,741 [main] INFO
com.yahoo.labs.samoa.evaluation.EvaluatorProcessor
(EvaluatorProcessor.java:86) - 1 seconds for 41920 instances
2015-09-01 22:22:17,760 [main] INFO
com.yahoo.labs.samoa.evaluation.EvaluatorProcessor
(EvaluatorProcessor.java:172) - evaluation instances = 41,920
classified instances = 41,920
classifications correct (percent) = 99.988
Kappa Statistic (percent) = -0.002
Kappa Temporal Statistic (percent) = 28.571
Exception in thread "main" java.lang.NullPointerException
at
com.yahoo.labs.samoa.learners.classifiers.trees.ModelAggregatorProcessor.process(ModelAggregatorProcessor.java:145)
at
com.yahoo.labs.samoa.topology.impl.SimpleProcessingItem.processEvent(SimpleProcessingItem.java:84)
at com.yahoo.labs.samoa.topology.impl.SimpleStream.put(SimpleStream.java:71)
at com.yahoo.labs.samoa.topology.impl.SimpleStream.put(SimpleStream.java:60)
at
com.yahoo.labs.samoa.learners.classifiers.trees.FilterProcessor.process(FilterProcessor.java:95)
at
com.yahoo.labs.samoa.topology.impl.SimpleProcessingItem.processEvent(SimpleProcessingItem.java:84)
at com.yahoo.labs.samoa.topology.impl.SimpleStream.put(SimpleStream.java:71)
at com.yahoo.labs.samoa.topology.impl.SimpleStream.put(SimpleStream.java:60)
at
com.yahoo.labs.samoa.topology.LocalEntranceProcessingItem.injectNextEvent(LocalEntranceProcessingItem.java:46)
at
com.yahoo.labs.samoa.topology.LocalEntranceProcessingItem.startSendingEvents(LocalEntranceProcessingItem.java:66)
at
com.yahoo.labs.samoa.topology.impl.SimpleTopology.run(SimpleTopology.java:42)
at
com.yahoo.labs.samoa.topology.impl.SimpleEngine.submitTopology(SimpleEngine.java:33)
at com.yahoo.labs.samoa.LocalDoTask.main(LocalDoTask.java:87)
We were able to track down the problem to the first instance that causes
it; the instance is on the 76426th line in kddcup_full.arff. The instance
is as follows:
1,tcp,smtp,SF,2252,331,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,1,7,0,0,0,0,1,0,1,5,216,1,0,0.2,0.01,0,0,0,0,normal
We haven’t noticed any differences between the problematic instance and the
other instances. Could you lead us to the root of the problem and could you
help us on how to overcome this problem?
As a workaround we’ve made the following addition to
ModelAggregatorProcessor.java
if (leafNode == null)
return false;
after the line
ActiveLearningNode leafNode = (ActiveLearningNode) foundNode.getNode();
Now, also VeriticalHoeffdingTree Classifier works fine on kddcup_full.arff. Is
this solution acceptable for the problem, what do you think?
Besides, we were wondering how we could fetch model contents such as
visiting nodes and node content etc.
Thanks for your help,
Respectfully,
Ercan Ozturk, Davut Deniz Yavuz, Gozde Boztepe, Sezin Gurkan
