Marc Glisse

> Can you explain why you you want `t[ep.astensor(i)]` instead of `t[ep.from_numpy(t, i)]` (i.e. why the second option is not good enough)? The one I would like most is `t[i]`....

With an example to demonstrate the issue: ```python import torch import eagerpy a = torch.tensor([0.], requires_grad=True) torch.norm(a, p=2).backward() print(a.grad) eagerpy.astensor(a).norms.l2().raw.backward() print(a.grad) ``` > tensor([0.]) > tensor([nan])

Hi, thanks for the reply. I use eagerpy so I can write the code only once and let it work with several frameworks. It is true that currently I mostly...

Thanks. As I mentioned in the report, I know I could filter, I just thought it would be nice if I didn't need to, if I could specify in just...

> Sorry I missed your point. Needrestart's config is pure perl so it should be possible to hack something. Maybe you could show the relevant line from your `/etc/default/grub` file....

Maybe at the same time, some methods still say they require `persistence()` when they should say `compute_persistence()`, probably they were in a branch when the others were changed.

You don't say very clearly what you tried. Did you run something like fix_inconsistencies_using_perturbation? Are there still inconsistencies afterwards? Does the simplicial complex you get not represent a triangulation?

Does the pykeops.torch version work, and does it impact performance?

https://github.com/getkeops/keops/issues/43 Pykeops support on windows doesn't seem to be there yet, so I think we need at least to keep a fallback.

> I used [the following script](https://gist.github.com/wreise/3e1d86c68377668ef4ff989f43caba41). IIUC this benchmark uses points below the diagonal, which isn't supported, so it would be safer without those. Also, it contains many points far...