I also saw openai ppl playing with tf. I wonder how/if caffe and torch will try to keep their usershttps://twitter.com/demishassabis/status/726079159051649024 …
-
@fchollet yeah i do that too hahaha keras privilege. That is why I thought a numpy/autodiff backend could be useful: debugging!
@edersantana very interesting idea but would require np support for a bunch of complex stuff: pooling, conv, autodiff...
-
-
@fchollet can still be fast-ish, but would require at least the following external packages: numba and autodiff -
@edersantana@fchollet it's not that fast but there is an amazingly good auto diff package that wraps numpy: autograd
End of conversation
New conversation -
-
-
@fchollet@edersantana@F_Vaggi Chainer's xp backend also has tons of potential for python-autograd integration. There an issue somewhere... -
@kastnerkyle@fchollet@edersantana I didn't know that. Autograd is my go to when I am too lazy to calculate Jacobian by hand - Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.