Re: Important features for each fold

There is a flag in RandomForest ("-attribute-importance) that makes it
output a ranking of attributes along with importance scores. It is not
based on the mechanism you describe though: it uses mean impurity decrease
instead.

Cheers,
Eibe

On Wed, 15 Feb 2023 at 03:23, Michael Hall <[email protected]> wrote:

>
>
> > On Feb 14, 2023, at 1:58 AM, Eibe Frank <[email protected]> wrote:
> >
> > If you use the AttributeSelectedClassifier, you will be able to set
> whatever attribute selection scheme you want as an option in that
> classifier. The selected attributes would then be passed to ZeroR (if you
> set that as the base classifier in AttributeSelectedClassifier). The
> process would be pointless from a classification point-of-view, but you
> would see the selected attributes in the textual description that is
> generated.
> >
> > Cheers,
> > Eibe
> >
>
> Thanks. I hadn’t looked at the attribute selection for some time. That
> does give you a very good variety of selection criteria.
>
> My thought had been that If you get say accuracy, (I’m not sure this is
> included in the model,) then for each attribute doing RandomForest you
> could keep an array of the accuracies of the results for each fold that
> attribute was included in, using whatever selection criteria RandomForest
> uses (random?), and then rank importance according to which attributes had
> the highest mean of their arrays. So which contribute most to the best
> accuracy? Possibly that already corresponds to what some selection criteria
> is doing.
>
>
> _______________________________________________
> Wekalist mailing list -- [email protected]
> Send posts to [email protected]
> To unsubscribe send an email to [email protected]
> To subscribe, unsubscribe, etc., visit
> https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
> List etiquette:
> http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html
>

_______________________________________________
Wekalist mailing list -- [email protected]
Send posts to [email protected]
To unsubscribe send an email to [email protected]
To subscribe, unsubscribe, etc., visit https://list.waikato.ac.nz/postorius/lists/wekalist.list.waikato.ac.nz
List etiquette: http://www.cs.waikato.ac.nz/~ml/weka/mailinglist_etiquette.html