pycm

pycm copied to clipboard

Description

Add multilabel support

We thought about this option, but it's not easy, most of PyCM parameters are not extendable to multilabel mode. @alirezazolanvari what is your opinion ?

These should be good: Precision, recall and F1 score

And also: Micro F1-score Hamming loss Jaccard index

Interesting link - micro and macro averaging - https://archive.is/fiF2C

Does this ticket cover the case where you have something like this?

Actual ["LabelA","LabelB"]

Predicted ["LabelA","LabelC","LabelD"]

Does this ticket cover the case where you have something like this?

Actual ["LabelA","LabelB"]

Predicted ["LabelA","LabelC","LabelD"]

Hi @radiantone Yes, this issue talk about multi-label classification. For example :

>>> actual = [["LabelA","LabelB"],["LabelC","LabelB"],["LabelC"]]
>>> predict = [["LabelA","LabelB"],["LabelB"],["LabelA","LabelC","LabelD"]]

As I mentioned above, it's not easy, because most of PyCM parameters are not extendable to multi-label mode. We will work on it in the future.

Any alternatives that support Multi-labels classification report?

Any alternatives that support Multi-labels classification report?

You can use scikit-learn multilabel confusion matrix