I've implemented a new torch nn criterion. I've done all the necessary steps for integrating this into torch (writing a .c file in nn/generic, adding lines to nn/init.c and nn/init.lua, etc.). However, I'm having issues building torch using my new stuff.
Here's what I did:
1) fork torch/nn on github
2) remove my local torch/nn directory, which is a clone of mainline github torch/nn
3) git clone my personal nn fork so that the torch/nn path now contains my clone.
4) add my new code to this clone.
5) cd .. and run install.sh
It's not working, however, because the build uses "git submodule update" to switch to the remotes/origin/getParamsByDevice branch of nn (which doesn't exist in my fork).
I'm sure other people have had a similar issue, where they've developed new modules that include c code in generic/ and want to use it locally (e.g. for testing) before making a torch pull request. How shouId I do it?
Thanks,
David