how to split a dataset into training and validation set keeping ratio between classes?

Question

I have a multi class classification problem and my dataset is skewed, I have 100 instances of a particular class and say 10 of some different class, so I want to split my dataset keeping ratio between classes, if I have 100 instances of a particular class and I want 30% of records to go in the training set I want to have there 30 instances of my 100 record represented class and 3 instances of my 10 record represented class and so on.

Answer 1

You can use sklearn's StratifiedKFold, from the online docs:

Stratified K-Folds cross validation iterator

Provides train/test indices to split data in train test sets.

This cross-validation object is a variation of KFold that returns stratified folds. The folds are made by preserving the percentage of samples for each class.

>>> from sklearn import cross_validation
>>> X = np.array([[1, 2], [3, 4], [1, 2], [3, 4]])
>>> y = np.array([0, 0, 1, 1])
>>> skf = cross_validation.StratifiedKFold(y, n_folds=2)
>>> len(skf)
2
>>> print(skf)  
sklearn.cross_validation.StratifiedKFold(labels=[0 0 1 1], n_folds=2,
                                         shuffle=False, random_state=None)
>>> for train_index, test_index in skf:
...    print("TRAIN:", train_index, "TEST:", test_index)
...    X_train, X_test = X[train_index], X[test_index]
...    y_train, y_test = y[train_index], y[test_index]
TRAIN: [1 3] TEST: [0 2]
TRAIN: [0 2] TEST: [1 3]

This will preserve your class ratios so that the splits retain the class ratios, this will work fine with pandas dfs.

As suggested by @Ali_m you could use StratifiedShuffledSplit which accepts a split ratio param:

sss = StratifiedShuffleSplit(y, 3, test_size=0.7, random_state=0)

would produce a 70% split.