Slightly off topic spilling of a jar of cents:

How about a metaestimator that is initialized with a (fitted)
estimator and has an API identical to pure feature selectors? That
would be more in the spirit of our API î think, but it's not much less
complicated than the way Gael thinks is too sophisticated.

On the other hand, we all know that users don't read docs, and having
awesome tools such as IPython makes it easy to guess what to do based
on an API. Do you think it's (intuitive/natural/expected) to find
.transform on a LinearSVC or on a Tree? Could you guess what it does?
Is there where you'd look for such a feature?

Rambling even further, once you transform with such a fitted model,
you (obviously?) can't predict the transformed data, because it has a
different dimension. So when you do feature selection with
trees/forests, the sequence is slightly strange:

1) fit on full-width data
2) transform training and test set
3) re-fit on transformed training data
4) predict the test set

Of course it all makes perfect sense but it's one of those things that
set API alarm bells in my head when I encountered it.

Cheers,
Vlad



On Fri, May 17, 2013 at 2:37 PM, Gael Varoquaux
<[email protected]> wrote:
>> Given one of these estimators, a perceptron, say, we can use:
>
>> perceptron.to_selector()
>
>> which returns an estimator whose parameters are generic things for feature
>> selection by score: a minimum score, maximum score,
>
> I think that this is software sophistication that makes it harder to use
> for people who are not used to complex software construct (think the
> matlab 101 user), and I for this reason, am -1.
>
> ------------------------------------------------------------------------------
> AlienVault Unified Security Management (USM) platform delivers complete
> security visibility with the essential security capabilities. Easily and
> efficiently configure, manage, and operate all of your security controls
> from a single console and one unified framework. Download a free trial.
> http://p.sf.net/sfu/alienvault_d2d
> _______________________________________________
> Scikit-learn-general mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

------------------------------------------------------------------------------
AlienVault Unified Security Management (USM) platform delivers complete
security visibility with the essential security capabilities. Easily and
efficiently configure, manage, and operate all of your security controls
from a single console and one unified framework. Download a free trial.
http://p.sf.net/sfu/alienvault_d2d
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to