search

andabi / deep-voice-conversion

Deep neural networks for voice conversion (voice style transfer) in Tensorflow
MIT License
3.92k stars 843 forks source link

Logdir storage requirements #31

Closed vivek-sethia closed 6 years ago

vivek-sethia commented 6 years ago

I am trying to run the two networks, and when I ran the two scripts script, the logdir folder used almost 40gb of my disk space and I had to terminate since I did not have much space left of my AWS instance. Any ideas on how to tackle this?

I have reduced to number of epochs for both the script to 500 only.

0i0 commented 6 years ago

not sure if its the right way but maybe change line 71: in train1.py to tf.train.Saver().save(sess, '{}/epoch_latest_step_latset'.format(logdir, epoch, gs)) so it will save only the latest model

vivek-sethia commented 6 years ago

Thanks for the reply. I solved it by setting the saver outside the loop in this way 

    saver = tf.train.Saver(max_to_keep=2)

and then using inside the loop like this :

    saver.save(sess, '{}/epoch_{}_step_{}'.format(logdir2, epoch, gs))
0i0 commented 6 years ago

that's probably better have you manged to run train2?

vivek-sethia commented 6 years ago

@0i0 I have managed to run train2 but olny with 500 epochs and the results are not good as expected. What about you?

0i0 commented 6 years ago

same here. 2000 epochs but with a different dataset