Add ROC curve for binary classification report

[pplonski]

It will be fantastic to add ROC curve in a README.md report file for the model in a binary classification task.

[diogosilva30]

Like this?

[diogosilva30]

This could also be extended to Multiclass classification, using a OneVsAll approach.

[pplonski]

Looks good!

• what about classes names, if there are no 0,1 class but coded categorically, can we add information which category is positive and which negative?
• It should be extended to multiclass.

I remember that @shahules786 was working on this issue. But the blocker was categorical class names.

[pplonski]

How would look the plot if there will be ROC from all data, not a mean?

From what I remember at the end of the training the reported AUC is computed from all available samples using out-of-folds predictions (in case of CV). So it will be good to have on the plot the ROC from all samples, instead of mean - to have consistency in results in the Markdown report and chart.

[shahules786]

@spamz23 There is an issue which is blocking further evaluation in working on multi class classification, maybe you can add it for binary classification. What say @pplonski ?

[pplonski]

For sure binary class and multi-class issues are connected. If there will be a way to solve it for any case, the other should be straight forward.

[shahules786]

Is this still on? If you want I can work on this. @pplonski

[pplonski]

Yes, it is available, I assigned you to this issue. Thank you!

[LukasKaspras]

@shahules786 Just saw that this issue is still open; can I support you in any way?