Benjamin Bossan

I would not try it :) Anyway, do you see a working solution for this?

So much for not breaking code in [this PR](https://github.com/dnouri/nolearn/pull/295) :) For those who use this snippet, in the part shown below, change `'_output_layer'` to `'_output_layers'`: ``` for attr in (...

Theano is the only backend for nolearn, whereas keras has several backends. This is why it makes sense to install the backend with nolearn but not with keras. > which...

I don't think you understand, Lasagne and Theano are not separate backends, Lasagne strictly depends on Theano. > 0.8 version of Theano This is the version required by Lasagne, a...

> Does it make sense to just pass it on and do the same thing as with slices? The problem here is the following: Say you have two values, np.zeros(10)...

I believe it would make sense as completely optional dependency.

How about `push` just forwarding all `kwargs` to `upload_folder`?

I just saw that `upload_folder` had a lot of arguments set to `None` so that seemed more important to me. Alternatively, we could accept two dicts as arguments, one passed...

I see your point. In the end, when providing an abstraction like `push`, we have to balance convenience and flexibility. If we're unsure, we can proceed by only adding `private`...

Once those are done, we should an entry for sklearn [here](https://huggingface.co/docs/hub/models-libraries).