F1 Score, Recall, Precision + log loss and rocauc + EvaluatePrequentialDelayedRegression

[danielnowakassis]

Fixes F1 Score Recall and Precision handling:

• Deals with classes not yet predicted (that might result in ? and should be treated as zero cases)
• Differentiate macro (per class and shows when per class option is activated) and micro
• results match sklearn

Adds log loss and roc auc to BasicClassificationPerformanceEvaluator

• adds log loss (new metric), matches sklearn
• roc auc is borrowed from Dariusz Brzezinski implementation in MOA.

Adds EvaluatePrequentialDelayedRegression

[hmgomes]

Thanks @danielnowakassis
I assigned myself to review it, should take 1 to 2 weeks from now due to workload

[danielnowakassis]

Hi @hmgomes.

Ok, thank you.

I'm just fixing some details in my next pr, will do it soon.

My recommendation for priority order:

1 - Metrics and delayed regression (most simple commit)
2 - incremental hyperparameter tuning algorithms (5 new algorithms)
3 - iPFI