When instantiating trainer guess which label is positive class for binary classifications

[Aylr]

Using the model_eval.gues_binary_positive_label() method.

Randomzied searches using roc_auc sometimes fail since there is no positive class label specified. I'm uncertain how to roll this down through the randomized search call...

One potential idea that I don't like is to guess the label, display it to the user and change that column to 1/0 according to that guess. Then the need to pass this down through the call stack to the roc_auc() function isn't required, though it could potentially obfuscate the data.

[Aylr]

@me this might be fixed in multiclass pre-release