Datasets with multiple targets

[hayesall]

The pos/neg/facts setup feels limiting when a dataset has multiple potential targets.

Consider imdb—where female_gender and workedunder are common targets.

There's usually a target in mind, but you can setup the problem with any target you like.

---

Side Note:

The following setup prescribes predicting y using X.

from sklearn.datasets import load_breast_cancer

X, y = load_breast_cancer(return_X_y=True)

But the vectors can be re-ordered to to predict the third column X[:, 3] after substituting y with the third column.