Orange Forum • View topic - bug in orngStat.AUC?

bug in orngStat.AUC?

Report bugs (or imagined bugs).
(Archived/read-only, please use our ticketing system for reporting bugs and their discussion.)
Forum rules
Archived/read-only, please use our ticketing system for reporting bugs and their discussion.

bug in orngStat.AUC?

Postby john » Tue Jan 25, 2005 20:28

This code works fine:

bayes = orange.BayesLearner()
data = orange.ExampleTable("voting")
results = orngTest.crossValidation([bayes], data, folds=10)
auc = orngStat.AUC(results)

but when I replace the 10-fold CV with leave one out:

bayes = orange.BayesLearner()
data = orange.ExampleTable("voting")
results = orngTest.leaveOneOut([bayes], data)
auc = orngStat.AUC(results)

then the call to orngStat.AUC raises an exception in checkNonZero in orngStat. The same exception occurs if I use

bayes = orange.BayesLearner()
data = orange.ExampleTable("voting")
results = orngTest.crossValidation([bayes], data, folds=len(data))
auc = orngStat.AUC(results)

Any ideas?

Postby Blaz » Tue Feb 08, 2005 20:50

Fixed. We needed to change computeCDT, which before returned 0 if any test set included only cases with same true labels (or included only one example, as with leave-one-out). New orngStat.py and corn.pyd are already on CVS, the two files should be in the distribution snapshot after midnight. Check also http://www.ailab.si/orange/doc/modules/orngStat.htm for new description of computeCDT.

Postby Janez » Wed Feb 09, 2005 0:04

John, if you're not in too much hurry it might be better if you wait for a few more days - we're just adding some new functionality to Orange which can make the snapshots rather unstable. We'll sure anounce it when it's done.


Return to Bugs