Vincent Lostanlen

Results 194 comments of Vincent Lostanlen

> I'm not following here. Are you saying that we would only require Q and J_fr to be kwarg-only by allowing J and shape to be positional? No, in my...

> I'm not following here. Are you saying that we would only require Q and J_fr to be kwarg-only by allowing J and shape to be positional? No, in my...

> I mean Q = 8 wouldn't be so bad, would it? As long as it's specified clearly in the docs. The problem is that a lot of users don't...

> I notice that the Keras frontend for the 1d scattering [hardcodes some input argument](https://github.com/kymatio/kymatio/blob/main/kymatio/scattering1d/frontend/keras_frontend.py#L15) This is a bug. It should be ```python self._max_order = max_order self._Q = Q ```...

@janden thanks. merge at your convenience!

Many thanks for opening this issue > why did we need contiguous tensors in the first place? https://stackoverflow.com/questions/69840389/what-functions-or-modules-require-contiguous-input/ "it seems that view is the only function that explicitly causes an...

Yes. This is a backwards compatible change (from the standpoint of frontends) so it can happen right now without deprecation.

What's the status with skcuda? Is it officially supported / deprecated / dropped? Do we test it?

Since we still have an skcuda backend, then maybe this (post-#972) is working as intended? @janden We have contiguous tensors all the way

that's really too bad! any idea what's going on? maybe related: #637