These are chat archives for dmlc/mxnet

13th
May 2017
Hanns Holger Rutz
@Sciss
May 13 2017 17:20
Hey there... I'm trying to understand which are my best options implementing something like WaveNet in Scala. So I found MxNet. The main problem I have is my Python is weak, and I don't know Tensorflow, Keras, or Torch. I'm fairly versed in signal processing, though. I would need to understand how to implement something like dilated convolution layers (https://github.com/ibab/tensorflow-wavenet/blob/12dc5df26b79c8b5ddaf608138129b328c12ded7/wavenet/model.py). Thanks for any comments.
I'm using a laptop with Core i7 quad, unfortunately no Nvidia, so needs to run on CPU, as parallelised as possible. Tensorflow implementation is extremely slow on this machine, making one epoch every half minute. I want to change the input signal model, and I hope I can speed things up somewhere by a magnitude.