regression_housing_2.py

def split(train_dataset):
    '''
    Shuffle data and split into 3 datasets
        1. Training - 60%
        2. Validation - 20%
        3. Testing - 20%
    '''
    # Shuffle data
    train_dataset = train_dataset.sample(frac=1)

    train, valid, test = np.split(train_dataset,
        [int(.6 * len(train_dataset)), int(.8 * len(train_dataset))])

    # Convert into numpy arrays
    x_train = train.drop(['SalePrice', 'Id'], axis=1).as_matrix().astype(np.float32)
    y_train = train['SalePrice'].as_matrix().astype(np.float32).reshape((np.shape(x_train)[0], 1))
    x_test = test.drop(['SalePrice', 'Id'], axis=1).as_matrix().astype(np.float32)
    y_test = test['SalePrice'].as_matrix().astype(np.float32).reshape((np.shape(x_test)[0], 1))
    x_valid = valid.drop(['SalePrice', 'Id'], axis=1).as_matrix().astype(np.float32)
    y_valid = valid['SalePrice'].as_matrix().astype(np.float32).reshape((np.shape(x_valid)[0], 1))

    return x_train, y_train, x_test, y_test, x_valid, y_valid