2013/1/10 Gael Varoquaux <gael.varoqu...@normalesup.org>:
> On Thu, Jan 10, 2013 at 03:57:23PM +1100, Juan Nunez-Iglesias wrote:
>> More precisely, I think David wants a function that will take a set of RFs 
>> and
>> return a new classifier object that does all the weighted averaging Andy
>> suggested for you transparently. And the answer is no, sklearn doesn't have
>> such a function. =)
>
> But it's fairly trivial to write an object that does that.

I think such a new method would be interesting as it could also be of
interest for linear models.

>>> clf1 = Perceptron().fit(X_bigdata_1, y_bigdata_1)

On another machine

>>> clf2 = Perceptron().fit(X_bigdata_2, y_bigdata_2)

Then averaging the models:

>>> clf3 = clf2.combine(clf1, copy=True)

A type error would be raised if clf1 is of a type that could not be
combined with clf2 (for instance if combining a linear model with a
RF).

Naive bayes models could also probably be extended to support such a
combine method if they would also store the `n_samples` of the
training set encountered at fit time as a new fitted attribute.

Off course the exact semantics of such a "combiner" method would be
model specific but the global expectation would be that combining two
compatible models trained on distinct shards of the data would result
in a model that generalize better than the original models even if not
always strictly equivalent to fit a unique model on the concatenated
data.

--
Olivier
http://twitter.com/ogrisel - http://github.com/ogrisel

------------------------------------------------------------------------------
Master Visual Studio, SharePoint, SQL, ASP.NET, C# 2012, HTML5, CSS,
MVC, Windows 8 Apps, JavaScript and much more. Keep your skills current
with LearnDevNow - 3,200 step-by-step video tutorials by Microsoft
MVPs and experts. ON SALE this month only -- learn more at:
http://p.sf.net/sfu/learnmore_122712
_______________________________________________
Scikit-learn-general mailing list
Scikit-learn-general@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to