TYP: get_reverse_indexer, get_group_index_sorter #40476
Conversation
jreback
added
the
Algos
|
any perf changes? |
might be some small gain on 32 bit, nothing on 64. |
pandas/core/indexes/base.py
Outdated
| new_codes = [lab[mask] for lab in new_codes] | ||
| left_indexer = left_indexer[mask] | ||
| # error: Value of type "Optional[ndarray]" is not indexable | ||
| left_indexer = left_indexer[mask] # type: ignore[index] |
There was a problem hiding this comment.
np.arange resolves to Any. https://github.com/numpy/numpy/blob/main/numpy/__init__.pyi
as a result left_indexer is not narrowed with left_indexer = np.arange(len(left), dtype=np.intp) (L4215 after merge master)
so I think in this case better to cast than add an ignore. (We have warn_redundant_casts = True in setup.cfg) so should be aware when the cast is no longer required)
| # missing values are placed first; drop them! | ||
| left_indexer = left_indexer[counts[0] :] | ||
| # error: Value of type "Optional[ndarray]" is not indexable | ||
| left_indexer = left_indexer[counts[0] :] # type: ignore[index] |
There was a problem hiding this comment.
you could maybe leave this one as a "fix later" ignore instead of a cast as we can fix ourselves by typing libalgos.groupsort_indexer (i.e. adding algos.pyi)
simonjayhawkins
left a comment
simonjayhawkins
left a comment
There was a problem hiding this comment.
Thanks @jbrockmendel for the updates
|
yeah will look a bit later |
3 participants
nailing down all the "np.intp"s bit by little bit.
AFAICT we're not going to be able to annotate ndarrays with ndim/dtype any time soon. Might be worth looking into https://pypi.org/project/nptyping/