Proposed feature

I have often had to use interpolation to plot the mean ROC or PR curves of several classifiers, as the number of thresholds can be distinct for each und thus their TPR, FPR, precision, and recall lengths vary. I thought it would be convenient to add a function argument "interp_dim" to the functions "roc_curve" and "precision_recall_curve" so the user can ensure that all classifier metrics have the same lengths, which would make it easier to plot the average.

Proposed solution

Perform numpy interpolation on TPR, FPR, precision and recall arrays to ensure they have the same length (as specified by the "interp_dim" function argument), which simplifies the computation of the mean.

I opened this issue to see if this idea receives support from the community before making a pull request.