doesn't it work for multi class classification. ?

[rajeshkalakoti]

Hello, I would like implement this library for multi class classification. could you please give an example for multi-class classification.

[olivertomic]

Hi! Unfortunately, there is no implementation yet for multiclass classification, but we will work on that problem in the near future. When working with a multiclass classification problem, we have a One-versus-All situation where we have as many classifiers as there are classes. Say we have 5 classes, then 5 models need to be trained, each trying to separate one specific class versus the rest of the classes. This implies that we get 5 weight vectors (not only one as in binary classification problems).

While this is very easily done in scikit-learn, we need to test out how we are going to handle and evaluate feature selections by each of the five models and decide on some sort of information summary that will be used for making the decision on which feature to select and which not. One of our master students will be working on this problem and if we find a good way to measure stability across many weight vectors, we will of course implement this approach into RENT. Until then, we can only offer feature selection for binary classification problems.

[rajeshkalakoti]

I understood. it's kind of pain staking trail and error procedure while training the multiple models and multiple classes. Thank you for the reply.