SOMA API v1 experiments [WIP/RFC]

[johnkerl]

This is some very, very rough initial work to get a first pass at looking for implementation-level gotchas with the nascent v1 SOMA-API spec.

• Basic walk-throughs:
  • See tests/test_v1_* at https://github.com/single-cell-data/TileDB-SingleCell/tree/kerl/v1-api-spec/apis/python/tests
    • Primarily https://github.com/single-cell-data/TileDB-SingleCell/blob/kerl/v1-api-spec/apis/python/tests/test_v1_soma_experiment_basic.py
  • See also the notebook https://cloud.tiledb.com/notebooks/details/johnkerl-tiledb/030375f5-2002-4d97-9b2a-0db0e810622b/preview
• In-package code status:
  • Basic obs, var, X have ingest-from-arrow working
  • Not all read methods exist yet; all of ReadResult is not implemented yet
  • get_schema (TileDB schema to Arrow schema) is not implemented yet
  • delete is not implemented yet
• Status with dependencies:
  • Indexing on string columns currently fails due to Unicode-vs-ASCII issues which were dealt with in v0 but are not yet dealt with in v1
  • When SOMADataFrame is stored as dense (not the case at the moment, because of these things):
    • exp.obs.read(ids=[0,2,4]) (discontiguous) SEGVs in core
    • This is done using the tiledb df accessor
    • We really want the multi_index accessor but we get the runtime error raise TileDBError("return_arrow=True requires .df indexer")` (@ihnorton FYI)
• Much more to do. Just a shared progress-point before I take a week of PTO.

[johnkerl]

Superseded by #282