Giovanni Palla

this is good test to take inspiration from @LLehner https://github.com/scverse/squidpy/blob/e561bc468cbf745e7ee2010b74824142625391a4/tests/graph/test_ppatterns.py#L16-L46

Hi @fritzo , I tried the first solution (`torch.set_default_tensor_type(torch.cuda.FloatTensor)`) and it doesn't seem to work. I get this error when using `torch.DataLoaders` and `pytroch_lightning.LightningDataModule` ``` RuntimeError: Expected a 'cuda' device...

thanks for the very prompt reply @fritzo , I was able to send some flows to `cuda` with `self.to(device)`. E.g. ```python AffineAutoregressive( AutoRegressiveNN( latent_dim, [hidden_units for _ in range(n_hidden)], skip_connections=True,...

Hi all, jumping on this thread a bit late, would be indeed super cool to have cellxgene to visualize visium data! Probably you solved issues already, but re `sc.read_visium`, indeed...

Hi @MaximilianLombardo , really cool prototype! I'd have couple of small comments: - from the readme, it seems like there is no use of the `spot_diameter`, a value that, multiplied...

I'll reopen this cause I think it's quite relevant still and could be very straightforward to implement with [sklearn resample](https://scikit-learn.org/stable/modules/generated/sklearn.utils.resample.html) also, there is an entire package for subsampling strategies which...

back here reminding myself that this would be very useful feature to have...

So assuming that we are only interested in downsampling, then I'd say `NearMiss` and related are straightforward and scalable (just need to compute a kmeans whcih is really fast)

also, the fact that reshuflling is performed is not in docs and should be documented. @bio-la do you plan to work on this?

hi @pacificoceanmist , which scanpy version are you using, and could you update to latest version ?