PERF: get_block_type

[lukemanley]

• closes PERF: get_block_type heavy use could benefit performance improvements #48212
• Tests added and passed if fixing a bug or adding a new feature
• All code checks passed.
• Added type annotations to new arguments/methods/functions.
• Added an entry in the latest doc/source/whatsnew/vX.X.X.rst file if fixing a bug or adding a new feature.

cc @jbrockmendel - this may partly close #48212, however, I suspect the OP was referring to non-EA's given the old version of pandas.

Performance improvement is mostly for EA's where the .kind call can be a bottleneck.

import pyarrow as pa
import pandas as pd
from pandas.core.internals.blocks import get_block_type

%timeit get_block_type(pd.ArrowDtype(pa.float64()))
# 3.51 µs ± 440 ns per loop (mean ± std. dev. of 7 runs, 100,000 loops each)    <- main
# 740 ns ± 5.19 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)  <- PR

%timeit get_block_type(pd.Float64Dtype())
# 1.3 µs ± 23.2 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)  <- main
# 289 ns ± 2.3 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)   <- PR

[jbrockmendel]

LGTM

[jbrockmendel]

we can improve a little bit here by checking kind in "fciub" instead of the list

[lukemanley]

updated

[jbrockmendel]

i think the SparseDtype check may no longer be needed

[lukemanley]

removed

[lukemanley]

your suggested updates give a bit of an improvement to non-EA's as well:

import numpy as np
from pandas.core.internals.blocks import get_block_type

%timeit get_block_type(np.dtype('float64'))

# 724 ns ± 59.4 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)  -> main
# 590 ns ± 30 ns per loop (mean ± std. dev. of 7 runs, 1,000,000 loops each)    -> PR

[jbrockmendel]

ping on green

[lukemanley]

ping on green

green - thanks

[jbrockmendel]

thanks @lukemanley