This script splits data loaded via GraphLab with two methods: SFrame.random_split and sklearn.cross_validation.train_test_split. In the end, present the number of different rows between each divided array.

test_sklearn_and_graphlab_split.py

#!/usr/bin/python

import graphlab
import numpy as np
from sklearn.cross_validation import train_test_split

def main():
    sales = graphlab.SFrame('kc_house_data.gl')
    train_data_graphlab, test_data_graphlab = sales.random_split(0.8, seed=0)
    input_graphlab = train_data_graphlab['sqft_living']
    output_graphlab = train_data_graphlab['price']

    data = sales['sqft_living', 'price'].to_numpy()
    train_data_sklearn, test_data_sklearn = train_test_split(data, test_size=0.2, random_state=0)
    input_sklearn = train_data_sklearn[:, 0]
    output_sklearn = train_data_sklearn[:, 1]

    graphlab_size = input_graphlab.size()
    sklearn_size = len(input_sklearn)
    print 'Size graphlab:', graphlab_size
    print 'Size sklearn:', sklearn_size

    max_size = graphlab_size if graphlab_size > sklearn_size else sklearn_size
    min_size = graphlab_size if graphlab_size < sklearn_size else sklearn_size

    diff = 0

    for i in range(min_size):
        if input_graphlab[i] != input_sklearn[i] or output_graphlab[i] != output_sklearn[i]:
            diff += 1

    print 'Number of different rows: {}/{}'.format(diff, max_size)

if __name__ == '__main__':
    main()