Possibility to do stratification with multi-output multi-class (multi-target) data

[bundit786]

Hi,
I have a multi-output multi-class (multi-target) dataset and would like to do data stratification before applying a learning algorithm. Using iterative_train_test_split from skmultilearn library (```
from skmultilearn.model_selection import iterative_train_test_split
x_train, y_train, x_test, y_test = iterative_train_test_split(x, y, test_size = 0.1)

Thank you.

[trent-b]

Hi, unfortunately this package is not currently designed to handle multiclass-multioutput classification. I wonder though if you could maybe one-hot encode your targets, feed them into the stratifier, keep track of the indices of the instances, and then, once they are split into folds, convert them back into the original target values since you've retained the indices. Just a thought. I'm sorry I couldn't be of more help.

[bablf]

This does work but I needed to use the MultiLabelBinarizer

[dyhan316]

Thank you @trent-b and @bablf ! Could you share a sample code that uses https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.MultiLabelBinarizer.html ?