Alex Wiltschko

Cross-cite autograd

3

Would be nice in the README to point Python folks to [autograd](https://github.com/hips/autograd), and I'm sure they'd be pumped to point Lua folks here.

https://github.com/torch/image/issues/168

Patch Luarocks and support all Torch rocks

3

Should have a package for each of: https://github.com/torch/rocks

Ravi Symbolic

1

Curious if you've made any progress on the `symbolic` extension mentioned in a talk you've given?

Try Torch!

4

I'm very interested in the things that you learn here. If you need some help getting NumPy or Torch examples written, let me know.

Conditionals supported?

2

I see you have support for some control flow (for-loops), which is awesome. I've been poking around the source, and don't see support for if-statements, while loops, or try/catch statements....

Optimized hyperparameters on "small" model, achieves 87 perplexity in 1 hour.

10

Applied bayesian optimization (whetlab.com) to lower the perplexity of the "small" model (13 full epochs).

ARC Compatibility

7

Novocaine currently doesn't work out of the box with Automatic Reference Counting (ARC) enabled. This is something that it definitely should support in the future.

Roadmap?

3

Is this the official torch9 repo? If so, what's the roadmap for release and expected improvements over torch7?