Support Baysor>=7.0.0

[quentinblampey]

Hi @lguerard, I added a test to check if the cell IDs are string
In the new 0.7.0 version, the cell IDs are strings, and we need to use the obs_names instead of CellID to compare to the loom matrix

Let me know if this works for you!

[quentinblampey]

Also, for now, you'll see that it doesn't support 3D segmentation. It will require additional updates to support it

[lguerard]

Thanks a lot for looking into that !

And thank you for including the dask backend so make it easier to do parallelization. I just tried with write_patches_cells but I'm getting an error, I'm not familiar at all with dask so I'm not sure what it means..

RuntimeError: Can't redefine method: __getstate__ on class: __torch__.torch.utils.mkldnn.MkldnnBatchNorm (of Python compilation unit at: 000001A8C4E17ED0)

[quentinblampey]

Could you open an issue with a detailed error log?
Since I'll merge this PR soon, it will be better to keep track of this issue

[lguerard]

I'm reinstalling my environment right now, I'll create an issue if I encounter it again.

But anyway, you removed the settings file and the dask support for now no ?

[quentinblampey]

Yes, I was making these changes in the dev branch, but now I have moved everything to the sopa2 branch.
Since these changes will be quite big, I decided to keep it for a future sopa==2.0.0 release

So you can try the sopa2 branch to test it out. Let me know!

NB: I saved the old tutorial as old_api_usage.ipynb to ensure backward compatibility (the old way to do it should still work), but you can try the api_usage.ipynb notebook (which is still in progress)

[quentinblampey]

Merging this PR, thanks @lguerard!
And please let me know regarding the issue with the dask parallelization (see my comment above)