Question about the OCRFsampling unsupervised method

[mlleberthe]

Hi,

My issue doesn't point out a specific problem, I just have one question about the OCRF unsupervised ROC and PR results (cf. bench_ocrf_roc_pr_unsupervised_factorized.png).

I presumed that this OCRF method refers to the "OCRFsampling" (Désir et al., 2013).

But I was wondering how these results generated with the bench_ocrf.py file, are supposed to be unsupervised since the fit_predict fonction needs the y_train and y_test (cf line 97 : scoring = model.fit_predict(X_train, y_train, X_test, y_test)).

Thank you for your help.

Best regards,

MlleBerthe

[ngoix]

Hi, I think this API it is due to the source code provided by the authors of OCRFsampling. I think y_train is yet not used during training. @RomainBrault can you confirm?

[drougui]

Hi, I also think that y_train is never used by https://github.com/ngoix/OCRF/tree/master/sklearn/ensemble/ocrf_open/ocrf_v0.6 but necessary to make the dataset have the proper structure with arff extension. https://github.com/ngoix/OCRF/blob/master/sklearn/ensemble/OCRF.py

[mlleberthe]

Ok ok, I see.

Thank you for the quick feedback :) :)

[RomainBrault]

Confirmed. It's never used.

We also ended up using X_train, y_train, X_test, y_test for supervised/unsupervised to help automatization of our tests, although as mentioned above y_train is not used in the unsup context.

2017-09-19 17:20 GMT+02:00 mlleberthe notifications@github.com:

Ok ok, I see.

Thank you for the quick feedback :) :)

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/ngoix/OCRF/issues/5#issuecomment-330573625, or mute the thread https://github.com/notifications/unsubscribe-auth/ABYz9-zhQdOxCUoARz9mDSxmZ5JVr4Njks5sj9sjgaJpZM4PceMz .

[mlleberthe]

Ok, thank you all for your answers :) 👍