Refactor snipping to enable trans, inter-arm, etc pileups

[sergpolly]

Minimally invasive snipping refactor to enable pileups in trans:

• user facing pileups behavior/API is unchanged !
• internal API changes:
  • view_df is no longer optional in the snipper-classes, it's a positional argument now
  • internal _pileup require explicit feature_type (bed/bedpe) parameter now - looks cleaner that way
  • internal _extract_stack is now more specialized - work for bed features only, extract_stack_pairwise deals with bedpe
• misc internal refactors:
  • all check removed from snipper-classes, moved to the user-facing pileups function
  • np.fmax trick for "on-diagonals" is now inside internal _pileups function
  • assign_view_auto -> combined_assignments_column=False -> no longer returns region column for BEDPE features
    • this was the first obstacle preventing trans pileups !
  • added more comments, where behavior seemed surprising
  • modified snipper classes:
    • min_diag doesn't require diagonal indicator matrix
    • expected - pre grouped and reindexed in the init
    • toeplitz expected works for arbitrary cis-region
    • removed old unused code for padding out of bound snippets - easier to read now
• Updated snipping tests:
  • added 2 tests where we actually compare values to naive clr.matrix.fetch results !
    • VERY important IMHO
    • used pytest-allclose for that https://cerfacs.fr/coop/pytest-allclose now in requirements-dev
  • fixed few windows/features - that were not view-annotated, but advertised as such
  • used pytest fixtures to define clr,exp,view_df that we reuse in most tests

TODO

• run flake/ruff/black on code
  • existing .flake8 -> [ruff]-section of pyproject.toml using flake8-to-ruff, ran run format on files modified by the PR
• add feature_type to user-API + default_cols pileup: bed vs. bedpe inference #427
• add arbitrary cis region tests

Future Notes

• the whole special treatment of BED-like/on-diagonal feature can probably be deprecated - i.e. on-diagonal should become a special case of bedpe, where region1==region2 - snipper-classes actually do that internally anyways - would yeild much cleaner code (same as in Refactoring snipping & pileups #227 (comment))
• the whole lo/hi business in the main-user facing function is not really used anywhere - snipper classes end up recalculating it anyways - so either cut it off, or actually make snipper-classes make use of lo/hi columns
• consider dealing with view-unannotated features separately e.g. in main pileups function

Misc Notes

• side effect of this refactor is that if one provides inter-arm features to the pileup together with just an intra-arm expected - the script will crash/raise KeyError

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB