Self-defined generator

generator.py

# definition of generator
def generator_from_ndarray(x, y, batch_size=32):
    
    number_of_batches = np.ceil(x.shape[0]/batch_size).astype(int)
    while True:
        for i in range(number_of_batches):
            yield x[i*batch_size:(i+1)*batch_size], y[i*batch_size:(i+1)*batch_size]

            
# x_train.shape = (5000, 100, 600, 1)
# y_train.shape = (5000, 1)
# x_valid.shape = (1000, 100, 600, 1)
# y_valid.shape = (1000, 1)

# the part of my fit_generator code        
history = model.fit_generator(
                          generator=generator_from_ndarray(x_train, y_train, 20),
                          steps_per_epoch=x_train.shape[0],
                          epochs=100, verbose=1, callbacks=callbacks,
                          validation_data=generator_from_ndarray(x_valid, y_valid, 20),
                          validation_steps=x_valid.shape[0], max_q_size=40,
                          workers=2, initial_epoch=init_epoch)