[FIXED] Train/Test/Validation stratified Set Splitting in Sklearn

Issue

I have a pandas dataframe in python that looks like this (but is more extended).

I want to split the data set into three sets: train (70%), test (10%) and validate (20%). Because there are some classes that have little objects, I want to also split them stratified. Is there a way (using scikit) to split them stratified in three sets?

This post shows how to do it without the stratify.

Solution

You could use train_test_split stratified twice:

non_validate_X, validate_X, non_validate_y, validate_y = train_test_split(X, y, stratify=y, test_size=0.2)
train_X, test_X, train_y, test_y = train_test_split(non_validate_X, non_validate_y,stratify=non_validate_y, test_size=1./8)

I use a split of 1./8 in the second one because 1./8 = 10%/(70% + 10%); the fraction of test in your train+test.

Answered By - Learning is a mess