Tooling: metrics evaluation

[hombit]

We need some function(s) to create a report about classification quality on validation subset (and maybe any other?).

Possible list (do not limit yourself =):

• Basic metrics: accuracy, recall, precision, log-loss, ROC AUC, F1, etc
• ROC plot
• Confusion matrix for recall or two matrices for both precision and recall
• Score histograms, and maybe calibrated plots