Cross validation with pure numpy

cross_validation.py

import numpy as np

np.random.seed(0)

def cross_validation(X, y, cv_size=0.1):
    """Run cross validation on a numpy ndarray and return corresponding indices as well

        @param: X data in the form of numpy ndarray
        @param: y labels in the form of numpy ndarray
        @param: cv_size size of the test set

    """
    data_size = X.shape[0]
    test_size = np.round(data_size * 0.1).astype(np.int32)
    total_idx = np.arange(0, data_size)
    test_idx = np.random.choice(total_idx, replace=False, size=(test_size))
    train_idx = np.setdiff1d(total_idx, test_idx)
  
    return X[train_idx, :], y[train_idx, :], train_idx, X[test_idx, :], y[test_idx, :], test_idx

The reason for using this instead of Sklearn's from sklearn.model_selection import train_test_split:

• avoid loading a library just for the sake of one method
• return the corresponding indices from each batch which comes in handy