Missing filter / featsel methods

[pat-s]

Filters

Pkg

No pkg

• AUC

• generic permutation

• univariate.model.score

stats

• anova

• kruskal

• linear.correlation

• rank.correlation

• variance

FSelector

Do we want to have these filters in again? Slow and Java problems..

FSelectorRcpp

• information.gain
• gain.ratio
• symmetrical.uncertainty

Learner integrated filters

• ranger.impurity

• ranger.permutation

• cforest.importance

Do we want to add the ramdomForest and randomForestSRC ones?

mRMRe

- [ ] mrmr -> slow and no support for classif tasks mlr-org/mlr#2604

praznik

• CMIM

• DISR

• JMI

• JMIM

• MIM

• MRMR

• NJMIM

care

• carscore

spFSR

Need to check.

Ensemble filters

• Min

• Mean

• Median

• Max

• Borda

• Borda-staircase

• Borda-power

[mllg]

For the filters: I'd start with stats / no pkg, then try to connect the modern filter packages (FSelectorRcpp and praznik).

We don't need 3 forest filters, we can solve this more generically by extending mlr3 learners with methods to extract feature scores.

[ja-thomas]

we can solve this more generically by extending mlr3 learners with methods to extract feature scores.

This is really important to be able to use all kinds of embedded feature selection directly by the learner.

[ja-thomas]

I don't really see a reason to use the Java FSelector package when there is FSelectoRcpp.

[pat-s]

I don't really see a reason to use the Java FSelector package when there is FSelectoRcpp.

The later does not have all filters of the former. See https://mlr.mlr-org.com/articles/tutorial/filter_methods.html.

[ja-thomas]

Well, with this argument we have to include all possible filters 😄

I would suggest we start without it, and if people complain/open issues we can still add them later.
Or are there any really important filters not yet in FSelectorRcpp?

[berndbischl]

there never is (should be) pressure to include everything, include what is most important

[pat-s]

My comment was more meant to be a comparison, not a statement that we should do it :)

[mllg]

NB: All learners which have some sort of "importance" are now supported via FilterVariableImportance.