On Fri, May 17, 2013 at 12:01 AM, Gael Varoquaux <
[email protected]> wrote:
>
> > For threshold, this additional parameter also makes it harder to adopt
> the
> > feature selection mixin.
> ...
>
The above should be object-level params, I believe.
>
So the Forest, LibLinear, Perceptron and SGD implementations should all
have an additional `transform_threshold` object-level parameter?
I am not firmly devoted to the following alternative, but I think it is
worth considering:
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, a limit on the number
of parameters returned from the max down... And it also has the underlying
estimator as a parameter so its parameters may be set in a grid search (see an
untested
implementation<https://github.com/jnothman/scikit-learn/blob/feat_sel_overhaul/sklearn/feature_selection/selector_mixin.py#L11>
).
What I like about this is that it doesn't clutter what are generally used
as classifiers or regressors with their secondary purpose as feature
selectors. Instead of automatically considering a Perceptron a Transformer,
it allows the Perceptron to be explicitly reinterpreted as one: a Pipeline
including a Perceptron among its transformers looks strange; a Pipeline
including a Perceptron().to_selector() makes more sense.
- Joel
------------------------------------------------------------------------------
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
