Orange Forum • View topic - Monitor progress of learning (a classifier/regression model)

Monitor progress of learning (a classifier/regression model)

A place to ask questions about methods in Orange and how they are used and other general support.

Monitor progress of learning (a classifier/regression model)

Postby lefterav » Sat Jul 14, 2012 19:16

Hi,

I have coded some Python code around Orange and I am running experiments, training over a big set of data (50,000 instances) with big feature vectors (260 features per instance). Some classifiers complete very shortly (Naive, kNN) but some others run for days or weeks, without reporting anything.

For the latter, I was wondering whether there is any existing functionality, where I can see the progress of the classifier as it is being learned. This would be really helpful for:
(a) seeing how fast particular models can be learned, and therefore see if the setting worths the effort
(b) avoid cases where something has crashed but the thread is still running

I though that some sort of verbose logging would provide this information, but I can't find any instructions for this

Re: Monitor progress of learning (a classifier/regression mo

Postby Ales » Mon Jul 16, 2012 9:41

Sorry no, there is no such functionality in general.

Which learners are causing you problems?

Re: Monitor progress of learning (a classifier/regression mo

Postby lefterav » Mon Jul 16, 2012 17:56

SVM has been running for a week now. I don't know what's its progress and what to expect, i.e. how long I should wait.

This is generally not very handy for obvious reasons. It is expected that scientific experiments may be running with a huge amount of data for a long time. How could I do a feature request?

Re: Monitor progress of learning (a classifier/regression mo

Postby Ales » Mon Jul 16, 2012 18:32

SVMLearner does have a 'verbose' option, although I see it is not documented (I will fix this).
Code: Select all
svm_l = Orange.classification.svm.SVMLearner(verbose=True)
turns on the native LibSVM output which is suppresed by default.

lefterav wrote:How could I do a feature request?
You can create a request ticker using our Trac: http://orange.biolab.si/trac/

Re: Monitor progress of learning (a classifier/regression mo

Postby lefterav » Tue Jul 17, 2012 12:38

Thank you very much. This is indeed a very interesting piece of information. Any information about other classifiers (C4.5, Tree, CN2) would also be appreciated

Re: Monitor progress of learning (a classifier/regression mo

Postby Ales » Wed Jul 18, 2012 9:48

CN2Learner has a `progress_callback` parameter
Code: Select all
def p(a1, *args):
    print a1
rl = Orange.classification.rules.CN2Learner(progress_callback=p)

The rest, I think don't support any progress monitoring.


Return to Questions & Support