Kyle Kastner

I am pretty much in agreement @benanne here - the only thing I would add is that it would be good to be sure that the trades made don't disallow...

You might need to closer match DCGAN's original setup - feeding the samples and data as separate minibatches to the discriminator is important. See the line here and just before...

My GPUs are still busy, but it is on my queue to try when I get one free (~2-3 more days I hope). One thing that might help is being...

AFAIK Theano and PyTorch are deterministic in their runs (from what ops I have used at least) given enough seeds are set and so forth. At very least, I have...

I was planning to side by side the RNN experiment, to avoid any issues with pooling, convolution and so on - backwards of max pooling being different in different frameworks...

This is also happening to me, as of the latest Theano

@mnick do you think adding support for lists as indices and values for sparse tensors is worthwhile? Basically something like ``` if type(subs) != np.ndarray: subs = np.array(subs) ``` and...

I am able to recreate the IPython error ``` UnicodeDecodeError: 'ascii' codec can't decode byte 0xd0 in position 272: ordinal not in range(128)) ``` using the aforementioned tmp.py, and In...

After making that change, I get an error in _both_ IPython and regular python - it looks like that pushes the error into mutiprocessing (which apparently doesn't support unicode in...

Still having trouble replicating in pure multiprocessing - even though it is running through the exact same line of multiprocessing (line 554), and args, kwargs are identical to the joblib...