Allow overriding of CSR accumulators with an IndexLike
#181
Conversation
|
Plz fix the formattings |
thetorpedodog
force-pushed
the
build-index-abstraction
branch
2 times, most recently
from
34306ab to
1d71cc1
Compare
johnkerl
changed the title
IndexLike
beroy
force-pushed
the
build-index-abstraction
branch
from
e808e51 to
4b18ba8
Compare
| @@ -26,7 +26,7 @@ def get_indexer( | |||
| """Something compatible with Pandas' Index.get_indexer method.""" | |||
|
|
|||
|
|
|||
| IndexFactory = Callable[[npt.NDArray[np.int64]], "IndexLike"] | |||
| IndexFactory = Callable[[npt.NDArray[np.int64], Optional[Any]], "IndexLike"] | |||
There was a problem hiding this comment.
The reason we have this take the NDArray as the only parameter is so that we can use the function pandas.Index as an index factory. With this added, pandas.Index no longer works, which is why the format check is now failing.
thetorpedodog
force-pushed
the
build-index-abstraction
branch
from
4b18ba8 to
1d71cc1
Compare
|
@thetorpedodog can you plz merge this as my C++ reindexer is dependent on this? |
|
I would like to wait until we get approval on the TileDB-SOMA side of the house. I am pretty confident that this is going to be it, but I don’t want to cut a release prematurely. It will be very quick—after merging this, we can immediately make the tag and everything will be ready in a matter of minutes. (There will be no rush to release TileDB-SOMA; this is fully compatible with existing TileDB-SOMA.) |
To support a custom indexer in TileDB-SOMA, this introduces a basic abstraction that is used to build the Indexer used by the CSR Accumulator. This can then be easily overridden by an implementation's custom subclass to swap that out without having to duplicate substantial parts of the code. The naming is, admittedly, not ideal; this is in part due to the naming of the things we're trying to abstract over (the Pandas `Index` type which has the `get_indexer` method).
beroy
force-pushed
the
build-index-abstraction
branch
from
1d71cc1 to
b47bbc8
Compare
|
We’re good to go! |
To support a custom indexer in TileDB-SOMA, this introduces a basic abstraction that is used to build the Indexer used by the CSR Accumulator. This can then be easily overridden by an implementation's custom subclass to swap that out without having to duplicate substantial parts of the code.
The naming is, admittedly, not ideal; this is in part due to the naming of the things we're trying to abstract over (the Pandas
Indextype which has theget_indexermethod).This is the somacore counterpart of single-cell-data/TileDB-SOMA#1728. It is entirely compatible with current code, though; once we’re positive that it fufills the needs of the tiledb-soma reindexer, we can merge it and release it and tiledbsoma installations will continue to work fine.