Use models.multibind instead of models.selexDataset in mubind-pipeline and make tests in mubind

[ilibarra]

• Define tests for Multibind (do not remove previous tests, but rather make new ones with the same style (also can group into a single script.
• update fit_model.py to use multibind for SELEX/PBM instead of If else. Probably, a parameter of Multibind self.datatype is enough for this change.

[ege-erdogan]

Regarding the second point, currently:

• train_iterative (in mubind prediction.py) called from fit_model initiates Multibind based on the passed dataset's class, also setting the datatype parameter.
• Multibind then initiates the correct module etc. based on the datatype param.

[ilibarra]

Solution can be

1. renaming SelexDataset to PBMDataset in this line of fit_model.py. This assumes that PBMDataset is working properly and it s the default solution.
   https://github.com/theislab/mubind-pipeline/blob/main/scripts/fit_model.py#L88

@johschnee may you please yes/no confirm that PBMDataset is the one to be used internally in prediction.py? Please write and/or DM us if more considerations have to be taken. Thank yo,
https://github.com/theislab/mubind/blob/main/mubind/tl/prediction.py#L349

[ilibarra]

• Test for PBMDataset in pytests is required.

[johschnee]

I unfortunately don't have access to the mubind-pipeline repository. But going by your description alone, I'd be careful about renaming. SelexDataset and PBMDataset have different attributes and these attributes are also used by the multibind model.

If you think the code in mubind-pipeline is relevant for a more datailed answer, it would be great if you could give me access.

[ilibarra]

@johschnee access to mubind-pipeline granted. My bad!

[johschnee]

Thank you, this indeed changes my answer. Yes, for PBM-Data the class PBMDataset is the one to be used internally. As Ege said, the datatype parameter is initialized based on the dataset's class.