相同的数据分为训练集、开发集和测试集

Question

我很难在以下函数中的每次迭代中获得相同的数据拆分？

def data(filename):

    X_train = data('train-images.gz')
    Y_train = data('train-labels.gz')
    X_test = data('t10k-images.gz')
    Y_test = data('t10k-labels.gz')

    X_train, X_devel = X_train[:, :-devel_size], X_train[:, -devel_size:]
    Y_train, Y_devel = Y_train[:-devel_size], Y_train[-devel_size:]

    return X_train, Y_train, X_devel, Y_devel, X_test, Y_test

调用上述函数时，如何使用相同的数据拆分来训练和验证？

原因是，我想用几种优化技术重新运行函数并比较准确性。

Answer 1

设置随机种子。

tf.random.set_seed(1)
np.random.seed(1)

相同的数据分为训练集、开发集和测试集

Same data split into training, dev and test set

python

numpy

mnist