验证数据可以是 tensorflow.keras 2.0 中的生成器（而训练数据是数组）吗？

Question

I want to train a model with tensorflow.keras . And I hope to do some other things between every training step, that's why I can not use fit_generator to train my model. In the other words, I hope to train model by looping fit function like follows:

training_data_generator = ...
validation_data_generator = ...
for i in range(steps_number):
    x, y = next(training_data_generator )
    net.fit(x, y)
    do_other_things_function()
    if steps_number%100==0:
        net.fit(x, y, validation_data = validation_data_generator)

But it failed. The reason may be that my training data is a numpy array in net.fit(x, y, validation_data = validation_data_generator) , but my validation data is a generator.

So my question is: how to use a numpy array as training data and generaor as validation data at the same time??

Answer 1

I recommend you turn your train data into a generator and use fit_generator() instead of fit()

There are more ways to tackle this, first, look at this question about how to change NumPy array to a generator, and if you need a more robust and detailed explanation look at this post.