autocorrect option for mcc

[ulf1]

it's not really a bug but might cause trouble. If all predictions are TRUE or all are FALSE then mcc will return NaN. It's kind of the main advantage of using Matthews correlations that the program would crash in case a predictive model generates only FALSE or TRUE.

ToDo: Add an autocorrect option that would impute 0

def mcc(x, axis=0, autocorrect=False):
...
    cm = confusion(x[:, i], x[:, j])
    r[i, j] = confusion_to_mcc(cm)
    if autocorrect: r[i, j] = impute_zero(r[i, j])
    ...

[ulf1]

481d01d