On Thu, Jul 12, 2012 at 10:03:41AM +0200, Andreas Müller wrote:
> What do you think? I'm not sure what would be the easiest way to accomplish
> this but I think it is worth investigating.

I agree.

> One possible way would be to build a normal model and  then have "apply_tree" 
> or "predict_tree"
> check the regularization conditions, i.e. make them stop at a certain depth.
> (or rather have separate methods that do that).

The way we have been doing regularization paths in the scikit so far is
to have a 'path' function that can fit the model on a full path and
return either a family of models, or, given test data, a list of test
errors. This path function is then called by a 'FoobarCV' object to
implement parameter selection.

I think that their would be value in formalizing a bit more this pattern,
as it pops up everywhere. It even goes beyond this: when doing parameter
selection on a model for which certain parameters have a path and others
do not, there would be value to use a combination of path and grid
search. Having a clear path API would help here.

I don't know if that answers your question, though.

Gael

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to