F-score results for Cross Validation on Multi-target data

Waikato / meka

Multi-label classifiers and evaluation procedures using the Weka machine learning framework.

http://waikato.github.io/meka/

GNU General Public License v3.0

200 stars 76 forks source link

F-score results for Cross Validation on Multi-target data #77

Open bundit786 opened 1 year ago

bundit786 commented 1 year ago

Hi Meka development team,

I have tried to evaluate a model for a multi-target dataset and would like to get the F-score result for individual target outputs with cross validation. Using Meka, it gives the accuracy but no F-score (Precision, Recall). Is there any way that I can produce the metric for this evaluation? I tried with Python's Sci-kit library but it doesn't support for multi-target data as well. Thanks in advance.

fracpete commented 1 year ago

F1 is not implemented for multi-target (I don't do multi-label/-target, so not sure whether it is even possible in general).

Here are the implementations:

multi-label stats get generated by MLEvalUtils.getMLStats
multi-target stats get generated by MLEvalUtils.getMTStats