clib.conversion._to_numpy: Add tests for pyarrow.array with pyarrow numeric types

[seisman]

Description of proposed changes

This PR adds tests for _to_numpy to check if it works for PyArrow arrays with pyarrow numeric dtypes.

PyArrow provides 11 numeric dtypes (https://arrow.apache.org/docs/python/api/datatypes.html): int8, int16, int32, int64, uint8, uint16, uint32, uint64, float16, float32, and float64. All can be converted to numpy dtype without issues.

Here is a minimal example to play with pyarrow types.

>>> import numpy as np
>>> import pyarrow as pa
>>> x = pa.array([1, 2, 3], type=pa.int16())
>>> x
<pyarrow.lib.Int16Array object at 0x7fc28a73fb80>
[
  1,
  2,
  3
]
>>> np.ascontiguousarray(x)
array([1, 2, 3], dtype=int16)

>>> x = pa.array([1, 2, None], type=pa.int16())
>>> x
<pyarrow.lib.Int16Array object at 0x7fc28a7ba620>
[
  1,
  2,
  null
]

>>> np.ascontiguousarray(x)
array([ 1.,  2., nan])

Please note that pa.float16 needs special handling:

>>> x = pa.array([1.0, 2.0, 3.0], type=pa.float16())
---------------------------------------------------------------------------
ArrowTypeError                            Traceback (most recent call last)
Cell In[9], line 1
----> 1 x = pa.array([1.0, 2.0, 3.0], type=pa.float16())

File ~/opt/miniforge/envs/pygmt/lib/python3.12/site-packages/pyarrow/array.pxi:370, in pyarrow.lib.array()

File ~/opt/miniforge/envs/pygmt/lib/python3.12/site-packages/pyarrow/array.pxi:42, in pyarrow.lib._sequence_to_array()

File ~/opt/miniforge/envs/pygmt/lib/python3.12/site-packages/pyarrow/error.pxi:155, in pyarrow.lib.pyarrow_internal_check_status()

File ~/opt/miniforge/envs/pygmt/lib/python3.12/site-packages/pyarrow/error.pxi:92, in pyarrow.lib.check_status()

ArrowTypeError: Expected np.float16 instance

>>> # The solution is from https://arrow.apache.org/docs/python/generated/pyarrow.float16.html#pyarrow.float16
>>> x = pa.array(np.array([1.0, 2.0, 3.0], dtype=np.float16), type=pa.float16())
>>> x
<pyarrow.lib.HalfFloatArray object at 0x7fc279308b20>
[
  15360,
  16384,
  16896
]
>>> # The numbers above look weird but when converting to numpy, they're correct
>>> np.ascontiguousarray(x)
array([1., 2., 3.], dtype=float16)

[seisman]

Please review this PR first before PRs #3584, #3585, #3596, since this PR is not affected by the behavior changes between pandas <2.2 and pandas >= 2.2. Then we can better know if workarounds in #3596 have side-effects.

[seisman]

I'm not sure why the Python 3.10 job keeps failing on macOS.

[weiji14]

Minor detail, but PyArrow calls them types instead of dtypes, e.g. at https://arrow.apache.org/docs/python/data.html#type-metadata. So should we call them types instead of dtypes?

Suggested change

	# PyArrow provides the following dtypes:
	#
	# - Numeric dtypes:
	# PyArrow provides the following types:
	#
	# - Numeric types:

[seisman]

Yes, I also noticed the difference. I'm OK with using type in the comments, but for the tests, currently, the test name is:

test_to_numpy_pyarrow_array_pyarrow_dtypes_numeric(dtype, expected_dtype):

Strictly speaking, it should be:

test_to_numpy_pyarrow_array_pyarrow_types_numeric(type, expected_type):

But I feel the minor changes in the tests are unnecessary.

[weiji14]

Yeah, I think they use 'type' in the comments, but use 'dtype' in the function name to be consistent with the other unit tests.

[seisman]

OK, in 111738c, I've updated 'dtype' to 'type' for comments about PyArrow and left the tests unchanged.