Refactoring snipping & pileups

[agalitsyna]

• The major problem with pileups is that the pileup function does not know about regions and takes them from the input features dataframe.

• The creation of the "region" column of features requires the user to run "assign_regions", which is a potentially obsolete and unnecessary step (see assign_regions needs to be fixed or removed #225).

• An alternative is to take the regions directly from snipper object, but the pileup takes only its methods. Should we replace passing two functions with a single snipper instead? If not, why?

[gfudenberg]

was this fixed @Phlya ? or should we convert to discussion?

[Phlya]

There will be a more user friendly API available imminently. So e.g. point 2 will be done internally. However the whole underlying Snipper API is not the most convenient and should be eventually replaced with perhaps dask? @golobor

Maybe we should convert the issue to a discussion as you suggest though! Since we need to think about low-level API design.

[gfudenberg]

update 9/14/2022:

• create a base class for snipper and re-use its methods.

  • this is because currently, adding modifications to some snipper (e.g., ExpectedSnipper) is already painful if you want to propagate the changes to other snippers.

• convert bed-like inputs to bedpe at the beginning of pileup and use that internally throughout. this can simplify code in multiple places. e.g.

  cooltools/cooltools/api/snipping.py

  Line 214 in 076a945

  region1 = region2 = support

  • this will also remove the need for the assign_view_auto, since we would always assign_view_paired