mlr-org / mlr3learners

Recommended learners for mlr3
https://mlr3learners.mlr-org.com
GNU Lesser General Public License v3.0
91 stars 15 forks source link

Does argument 'censored' make sense for our classif / regr NNet learners? #313

Open mb706 opened 1 month ago

mb706 commented 1 month ago

NNet doc about the 'censored' hyperparameter says

A variant on softmax, in which non-zero targets mean possible classes. Thus for softmax a row of (0, 1, 1) means one example each of classes 2 and 3, but for censored it means one example whose class is only known to be 2 or 3.

This sounds like it is about multilabel tasks, but our tasks always have a single label for each sample, so this should not apply, right? Also our regression NNet should not have this hyperparameter at all? I haven't read the docs thoroughly, though, so not sure.