Sander Dieleman
Sander Dieleman
> Why don't you cast the data _before_ turning it into a shared variable? I used `np.uint8` in Lasagne's MNIST example so they're GPU compatible. This carries over from the...
I will try to finish this soon, just a bit occupied with other things at the moment. I haven't abandoned this PR!
just f0k's comments that need to be addressed. I should have some time to sort this out soon, but feel free to do it if you want this to be...
sounds good to me.
That is worth considering, although I'm finding it hard to think of / explain the potential use cases (which we would have to do in the docs).
For someone (like me) whose experience with RNNs is limited, the first (current) approach does seem a lot more natural, but I appreciate that some might need / expect more...
The documentation does not say that you can pass an int though, only tuples of int/`None`. If there's a particular passage that lead you to believe that you can pass...
Looks like this has been outstanding in Theano for a while, so we probably shouldn't count on it getting fixed very soon. We could put your workaround in Lasagne I...
Not to my knowledge. If you have an implementation it would be a welcome addition to our Recipes repository! https://github.com/Lasagne/Recipes
This might be a bit tricky as it requires storing the initializers for future use, which is currently not done. So it's not a small change, probably. Is there a...