Feature: Support Multi-label classification outputs

[eddiebergman]

Feature

Having the library support Multi-label/mult-target outputs would be nice. As you mentioned, MultiOutputClassifier from sklearn could be used to adapt the library in a general case.

Benefit

Increased usability in different domains and datasets.

[pplonski]

@eddiebergman thank you for the feature request! Do you have example datasets that will be good for testing / benchmarking this functionality?

[eddiebergman]

Unfortunately, my use case is currently on non-public modified data.

For testing, sklearn.datasets.make_multilabel_classification (doc) could be used.

As for a benchmark I could point you to the follow datasets from OpenML, they are all tabular datasets:

• yeast - 14 classes, genetic
• emotions - 6 classes, feature extracted audio
• image - 5 classes, feature extracted images
• reuters - 7 classes, feature extracted text

... however I'm sure there are more standard benchmarks in that sense that should be considered and have prior performances for comparison.

[pplonski]

Great, thank you @eddiebergman :+1:

[AkshayNovacene]

Hello @pplonski I wanted to know if MlJar still doesn't support Multi-label classification?

[pplonski]

Hi @AkshayNovacene,

We dont support multi label classification. We will talk internally with @Bocianski and maybe we will add it soon, we will let you know.