Add stats.is_constant

[flying-sheep]

Fixes #28

I don‘t actually understand why the dask tests pass, but they do.

@ilan-gold @Intron7 can you help me understand why this works?

fast-array-utils/tests/test_stats.py

Lines 123 to 133 in 19a4187

	def test_is_constant_dask() -> None:
	"""Tests if is_constant works if each chunk is individually constant."""
	if TYPE_CHECKING:
	import dask.array.core as da
	else:
	import dask.array as da
	x_np = np.repeat(np.repeat(np.arange(4).reshape(2, 2), 2, axis=0), 2, axis=1)
	x: da.Array = da.from_array(x_np, (2, 2)) # type: ignore[no-untyped-call]
	result = stats.is_constant(x, axis=None).compute() # type: ignore[attr-defined]
	assert result is False

TODO:

• benchmarks
• enable doctests
• sparse-in-dask
• zarr/hdf5?

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.95%. Comparing base (ef6b847) to head (521fb63).
Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #21      +/-   ##
==========================================
+ Coverage   95.53%   96.95%   +1.41%     
==========================================
  Files           9       11       +2     
  Lines         112      164      +52     
==========================================
+ Hits          107      159      +52     
  Misses          5        5              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[codspeed-hq]

CodSpeed Performance Report

Merging #21 will not alter performance

_{Comparing pa/is-constant (521fb63) with main (ef6b847)}

Summary

✅ 34 untouched benchmarks
🆕 18 new benchmarks

Benchmarks breakdown

	Benchmark	BASE	HEAD	Change
🆕	test_stats_benchmark[0-numpy.ndarray-float32-is_constant]	N/A	230.1 µs	N/A
🆕	test_stats_benchmark[0-numpy.ndarray-float64-is_constant]	N/A	274.7 µs	N/A
🆕	test_stats_benchmark[0-scipy.sparse.csc_array-float32-is_constant]	N/A	93.5 µs	N/A
🆕	test_stats_benchmark[0-scipy.sparse.csc_array-float64-is_constant]	N/A	102.3 µs	N/A
🆕	test_stats_benchmark[0-scipy.sparse.csr_array-float32-is_constant]	N/A	567.4 µs	N/A
🆕	test_stats_benchmark[0-scipy.sparse.csr_array-float64-is_constant]	N/A	607.8 µs	N/A
🆕	test_stats_benchmark[1-numpy.ndarray-float32-is_constant]	N/A	228.6 µs	N/A
🆕	test_stats_benchmark[1-numpy.ndarray-float64-is_constant]	N/A	272 µs	N/A
🆕	test_stats_benchmark[1-scipy.sparse.csc_array-float32-is_constant]	N/A	566.6 µs	N/A
🆕	test_stats_benchmark[1-scipy.sparse.csc_array-float64-is_constant]	N/A	606.9 µs	N/A
🆕	test_stats_benchmark[1-scipy.sparse.csr_array-float32-is_constant]	N/A	488.1 µs	N/A
🆕	test_stats_benchmark[1-scipy.sparse.csr_array-float64-is_constant]	N/A	152.7 µs	N/A
🆕	test_stats_benchmark[None-numpy.ndarray-float32-is_constant]	N/A	109.4 µs	N/A
🆕	test_stats_benchmark[None-numpy.ndarray-float64-is_constant]	N/A	129.9 µs	N/A
🆕	test_stats_benchmark[None-scipy.sparse.csc_array-float32-is_constant]	N/A	111.4 µs	N/A
🆕	test_stats_benchmark[None-scipy.sparse.csc_array-float64-is_constant]	N/A	132.7 µs	N/A
🆕	test_stats_benchmark[None-scipy.sparse.csr_array-float32-is_constant]	N/A	111.1 µs	N/A
🆕	test_stats_benchmark[None-scipy.sparse.csr_array-float64-is_constant]	N/A	131.7 µs	N/A

[ilan-gold]

Good TODO

[ilan-gold]

I can look into why the above works

[flying-sheep]

OK, using asymmetric overlap this looks almost as efficient as it could be:

since it implies boundary="none" this means our example chunks become something like:

[[0, 0, 1],  | [[1, 1],
 [0, 0, 1],  |  [1, 1],
 [2, 2, 3]]  |  [3, 3]]
             |
-------------+---------
             |
[[2, 2, 3],  | [[3, 3],
 [2, 2, 3]]  |  [3, 3]]

[ilan-gold]

Should this test be renamed to something maybe more informative (test_dask_constant_blocks?), and then the other test be updated to use dask? Am I missing something or does the test_is_constant one skip dask?

[flying-sheep]

it doesn’t, this one just tests a specific possible failure in dask, so I should definitely rename it.