3

Theano and Torch both supports GPU calculations. My question is whether Theano or Torch have significant differences in:

  • performance
  • ease of use (assuming one knows the programming language)
  • libaray support for optimizaiton algorithms (e.g. pylearn2 for Theano)
  • parallelizm (regarding parallel calculation inside one PC and distributed calcualtions across computers... Theano is not fit for the latter)
  • hardware support (e.g. OpenCL support is quite young in Theano)

EDIT

Torch actually doesn't have symbolic differentiation, so no question which one to use for my own optimization problems (having no time to do the differentiation for each problem).

Also only Theano is symbolic/functional of the two, which I actually do enjoy. (Some might not.)

This reddit link was pretty helpful to learn about the differences.

Mark Horvath
  • 795
  • 1
  • 8
  • 9
  • This link sums it up pretty well, but I personally prefer Torch. You will need to know Lua and you can't be afraid of hacking with Torch.... However, aside from coding with Lua, it's way easier than Theano... and faster and has some of the best in the field updating the repositories... But I learned everything I know from LeCun using Torch, so I'm biased. http://fastml.com/torch-vs-theano/ Caffe is another (python based) option that I like. – justanotherbrain Jul 21 '15 at 04:17
  • @justanotherbrain, thanks fort pointing out other options too. While I had this feeling with more support and larger ecosystem for Python, I still wonder why DeepMind uses the Lua based option (where one probably has less research/plotting/etc libraries). – Mark Horvath Jul 21 '15 at 16:34
  • This might be a more suitable question for programmers over at Stack Overflow. – Gavin M. Jones Jul 23 '15 at 23:35

0 Answers0