Adam Paszke
Adam Paszke
@szagoruyko `nn.SpatialConvolution` now supports `:noBias()`
Hi Natalie, 1. Subsampling simply means that I select every e.g. 10th sample from the training set and discard all the rest. This allows to make all events shorter (in...
I can of course answer some more specific questions, but most of this patch is basically a tiny bit of [reverse mode AD code](https://rufflewind.com/2016-12-30/reverse-mode-automatic-differentiation), and a ton of defensive programming...
@abadams Thanks for the review! Great, it seems that we're on the same page! 😄 I agree that shifting the formulas to lhs isn't perfect, but it could easily be...
Yep, that's exactly what I've been thinking. Great that we're on the same page!
It depends on how you're counting the bit numbers. If you start from 0 then changing the bottom index from 0 to 1 is fine :) Anyway, it's a great...
Unfortunately this is not a `ghci` problem --- I'm getting those errors whenever I `cabal build` (I don't really care about `ghci` and only want to be able to build...
Ah I didn't realize that, although it should have been obvious in hindsight 😅 Makes sense. I wonder what did you have to change in the recipes to make them...
Nope, we're using Stack. You can find the relevant configuration [here](https://github.com/google-research/dex-lang) (in the `stack.yaml`, `stack-macos.yaml` and `dex.cabal` files).
You can see an example with `autograd.gradcheck` right above [this header](http://pytorch.org/docs/notes/extending.html?highlight=gradcheck#extending-torch-nn).