Le 25 mars 2012 12:16, Gilles Louppe <[email protected]> a écrit :
> Hi Olivier,
>
> The higher the number of estimators, the better. The more random the
> trees (e.g., the lower max_features), the more important it usually is
> to have a large forest to decrease the variance. To me, 10 is actually
> a very low default value. In my daily research, I deal with hundreds
> of trees. But yeah, it also takes longer.

Indeed. I think we should put some practical scales somewhere in the
doc, maybe something along the lines:

Depending on the max_depth of the trees and the size of the dataset:

- 10+ trees: a couple of seconds or minutes of sequential CPU time,
suitable for debugging
- 500+ trees: a couple of minutes or hours of sequential CPU time,
suitable for getting interesting results  (requires multi-core
computation in practice)
- 5000+ trees: a couple of hours or days of sequential CPU time,
suitable for getting appearing in the leaderboards of machine learning
challenges (requires distributed computation in practice)

> By the way I am curious, what kind of dataset are you testing those
> methods on? :)

I used 
http://scikit-learn.org/dev/datasets/index.html#the-olivetti-faces-dataset
which is stupid because of the large number of classes as explained by
Peter.

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

------------------------------------------------------------------------------
This SF email is sponsosred by:
Try Windows Azure free for 90 days Click Here 
http://p.sf.net/sfu/sfd2d-msazure
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to