These are chat archives for dmlc/mxnet

Mar 2016
Kota Weaver
Mar 14 2016 15:35
hi guys, would you say MXNet is stable enough/featureful enough to be production ready? i'm looking for a good C++ deep learning framework that i can use in production, and the notion of being able to do the research and training stages in other languages (such as Julia) really appeals to me.
Tianqi Chen
Mar 14 2016 20:00
it is used by several startups already
e.g. @winstywang is using it on their production
Kota Weaver
Mar 14 2016 20:11
@tqchen thanks :) yeah, i started looking at MXNet.cpp (actually, just submitted a quick patch/pull request, probably won't be accepted but that's okay).
@tqchen i was wondering though, what's the right way to concatenate two variables? ie, i have a network which has a CNN section and a fully connected section, and the two are combined at a later stage, before getting pushed into a set of fully connected layers. i know i probably want to flatten the CNN before combining, but is it just a matter of setting the outputs of the CNN and the MLP sections into a variable and adding them together or something?
thanks again for the help by the way
Kota Weaver
Mar 14 2016 20:43
oh, i guess there is a concat directly in mxnet's symbol library... dunno how i missed that earlier.