Add other accuracy metrics to explore plots

[holder66]

In addition to raw accuracy scores in the plots generated by explore, there should be an option to plot balanced accuracy or F1 scores using either macro or weighted averaging. For explore plots on datasets without continuous attributes, or when only a single binning value is used, one plot can show separate traces for each of the different accuracy scores.

[holder66]

I changed the metric for the accuracy by number of attributes plot, to show balanced accuracy instead. Not sure if F1 score here would add anything.

[holder66]

Unfortunately, there are two formulas for balanced accuracy, one for binary (2-class) problems, the other for multiclass problems. Currently, the graph for explore shows the multiclass balanced accuracy only.