Hi Christian.
I suggest you look into "Elements of Statistical Learning".
It is a great book about machine learning and it focuses quite a bit
on interpretation of learned models - the view taken is often that
you want to find cause and effect, not predict well on unseen data
(which they call "data mining applications" iirc). It is available as
pdf for free.

As far as I know, the easiest approach to understanding the features is feature selection.
That will tell you what features play a more important role and which don't.
Look into the user guide on feature selection <http://scikit-learn.org/dev/modules/feature_selection.html> to see what you can do with sklearn.

As I have no idea about feature selection, I'll stop here.
Gael and Alex are the experts in this field, I believe...

Cheers,
Andy

On 10/01/2012 09:49 PM, Christian Jauvin wrote:
Hi everyone,

I have this (rather vague) intuition that studying the "reasons" which
led a trained classifier to behave like it did on particular instances
of a problem might be a good way to increase its understanding. If you
have for instance a very imbalanced problem, it might be useful to
identify the few cases where a (trained) classifier answered right (in
terms of classification or probabilistic output) on the least likely
class, in order to determine which particular features have played a
positive role, and which haven't. The way I see it, this would be a
bit like "reverse engineering the features".

So my question: is there a mechanism or maybe an already existing
framework or theory for doing this? And would something approaching it
be possible currently with Sklearn?

Thanks,

Christian

------------------------------------------------------------------------------
Got visibility?
Most devs has no idea what their production app looks like.
Find out how fast your code is with AppDynamics Lite.
http://ad.doubleclick.net/clk;262219671;13503038;y?
http://info.appdynamics.com/FreeJavaPerformanceDownload.html
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

------------------------------------------------------------------------------
Got visibility?
Most devs has no idea what their production app looks like.
Find out how fast your code is with AppDynamics Lite.
http://ad.doubleclick.net/clk;262219671;13503038;y?
http://info.appdynamics.com/FreeJavaPerformanceDownload.html
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to