Justin Johnson

For more complex recurrence patterns like this, you may have better luck with [torch-rnnlib](https://github.com/facebookresearch/torch-rnnlib) or [Element-Research/rnn](https://github.com/Element-Research/rnn).

I think that this would be pretty difficult. One easy thing to implement would be rejection sampling: keep drawing random samples until you get one that satisfies your constraints. However...

An easy way to get a deep copy of anything in Torch is to serialize and deserialize the same object: ``` lua local model_copy = torch.deserialize(torch.serialize(model)) ```

Looks like you using a GTX 1080; for Pascal GPUs to work you'll need to install the CUDA 8.0 Release Candidate. After updating CUDA, you'll need to reinstall cutorch and...

Looks like you are running out of memory; the Quadro FX 580 you are using only has 512MB of VRAM. You can try a smaller `rnn_size` or `seq_length`.

To backprop on only a single hidden state you will need to construct a gradient tensor of shape N x T x H which is all zero except in the...

I don't quite understand exactly what you mean by masking vs zero padding; can you explain the difference?

You can see an example of using masks for minibatches by looking at the LanguageModel from DenseCap: https://github.com/jcjohnson/densecap/blob/master/densecap/LanguageModel.lua

That's a pretty big RNN! It's a bit slow but runs fine on my system. Maybe you are running out of memory and swapping to disk? That would slow things...

The .t7 checkpoint is 659MB, and takes maybe 10 seconds to save on my system. On Thu, Apr 7, 2016 at 3:38 PM, AlekzNet [email protected] wrote: > Justin, just curious,...