Lesson 9: neural-sr: train_on_batch doesn't use the gpu?


(Sven) #1

Hi,

I am just getting the neural super-resolution example to work and noticed that Keras is not using the GPU for running the loop around ``m_sr.train_on_batch([lr[:bs], hr[:bs]], targ)’’ in the train function. Still, when I setup the notebook on top it says I should be using the tensorflow backend:

Using TensorFlow backend.

def train(bs, niter=10):
    targ = np.zeros((bs, 1))
    bc = BcolzArrayIterator(arr_hr, arr_lr, batch_size=bs)
    for i in range(niter):
        hr,lr = next(bc)
        m_sr.train_on_batch([lr[:bs], hr[:bs]], targ)

Any idea why it is not running on the GPU? Do my batches maybe use too much memory so they cannot fit on the GPU?

Thanks!


#2

maybe you didn’t install the tensorflow-gpu version?


(Sven) #3

@justinho Yeah, you are right. I figured it out now, thanks!