Dataframe for non-matrix classification metrics

[fkromer]

It would be valuable to have some mapping functionality which takes y_true, y_pred, y_score, pred_decision (output of estimator.decision_function()), x (AUC only), y (AUC only) and converts it into a standardized pandas DataFrame ("non-matrix classification metric dataframe") which could look like follows for a single estimator:

metric	accuracy_score	auc	...	jaccard_score	zero_one_loss
estimator
LogisticRegression	float or int	float		float or [float, ..., float]	float or int

... could include as well:

• cohen_kappa_score (float)
• dcg_score (float)
• f1_score (float or [float, ..., float])
• hamming_loss (float or int)
• hinge_loss (float)
• log_loss (float)
• ...

In case an estimator does not support one or several metrics the cells could contain either NaN

A dataframe containing several estimators could look like e.g.:

metric	accuracy_score	auc	...	zero_one_loss
estimator
LogisticRegression	float or int	float		float or int
RidgeClassifier	float or int	float		float or int

[QuadV]

This sounds interesting. Could be a good addition to the repo. Can I take it up?