Convert pickled RegularGridInterpolator objects to numpy saved files

[scottstanie]

The implementation of RegularGridInterpolators changed after scipy 1.14. If you ran this, you'd get the error from #2 :

uvx --with scipy==1.15 ipython

In [1]: spline = pickle.load(Path("src/whirlwind/carballo-pdf-0-spline.pkl").open('rb'))

In [2]: spline((.1, .1, 10))
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
Cell In[8], line 1
----> 1 spline((.1, .1, 10))

File ~/.cache/uv/archive-v0/Zn_jXn_9BGPpxhZbmyQ53/lib/python3.13/site-packages/scipy/interpolate/_rgi.py:391, in RegularGridInterpolator.__call__(self, xi, method, nu)
    334 def __call__(self, xi, method=None, *, nu=None):
    335     """
    336     Interpolation at coordinates.
    337
   (...)    389     array([ 4.7, 24.3])
    390     """
--> 391     _spline = self._spline
    392     method = self.method if method is None else method
    393     is_method_changed = self.method != method

AttributeError: 'RegularGridInterpolator' object has no attribute '_spline'

This PR saves the underlying numpy arrays so that we can load it in newer scipy versions.

Closes #2

[gmgunter]

Thanks! Looks like a good fix to me. Glad to see that this tool is still getting some love.

[scottstanie]

Changed those two items, thought I had done something else wrong since a SNAPHU-like unit test (with unw = whirlwind.unwrap(igram, corr, nlooks=1.0) instead) lead to this:

but it's happening on the main branch, as well as a couple git commits back before i touched anything

[gmgunter]

Weird! Seems like it's an issue with the residue calculation -- nothing to do with the unwrapping algorithm itself. I don't think there should be any residues in that interferogram, but

import numpy as np
from whirlwind._lib import residue as get_residues

y, x = np.ogrid[-3:3:512j, -3:3:512j]
phase = np.pi * (x + y)
igram = np.exp(1j * phase)

wrapped_phase = np.angle(igram)
residue = get_residues(wrapped_phase)

num_negative_residues = np.sum(residue == -1)
num_positive_residues = np.sum(residue == 1)

print(f"{num_negative_residues:d}")  # prints '6'
print(f"{num_positive_residues:d}")  # prints '6'

Thanks for the report! It's clearly unrelated to this PR. I'll see if I can fix it in a separate PR.

[gmgunter]

Using numpy=2.3.5 and the latest commit

$ git rev-parse HEAD
46560ee9b55e331be1b82b2bf8e831f50ced7d32

I'm getting the following error:

Traceback (most recent call last):
  File "/home/ggunter/Projects/whirlwind/./test.py", line 12, in <module>
    unw = ww.unwrap(igram, corr, nlooks=1.0)
  File "/home/ggunter/miniforge3/envs/whirlwind/lib/python3.14/site-packages/whirlwind/_unwrap.py", line 28, in unwrap
    cost = compute_carballo_costs(igram, corr, nlooks, mask)
  File "/home/ggunter/miniforge3/envs/whirlwind/lib/python3.14/site-packages/whirlwind/_cost.py", line 65, in compute_carballo_costs
    spline_pdf0, spline_pdf1 = load_carballo_pdf_splines()
                               ~~~~~~~~~~~~~~~~~~~~~~~~~^^
  File "/home/ggunter/miniforge3/envs/whirlwind/lib/python3.14/site-packages/whirlwind/_cost.py", line 50, in load_carballo_pdf_splines
    spline_pdf0 = _load_rgi(p)
  File "/home/ggunter/miniforge3/envs/whirlwind/lib/python3.14/site-packages/whirlwind/_cost.py", line 37, in _load_rgi
    points=tuple(data["grid"]),
                 ~~~~^^^^^^^^
  File "/home/ggunter/miniforge3/envs/whirlwind/lib/python3.14/site-packages/numpy/lib/_npyio_impl.py", line 257, in __getitem__
    return format.read_array(
           ~~~~~~~~~~~~~~~~~^
        bytes,
        ^^^^^^
    ...<2 lines>...
        max_header_size=self.max_header_size
        ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
    )
    ^
  File "/home/ggunter/miniforge3/envs/whirlwind/lib/python3.14/site-packages/numpy/lib/_format_impl.py", line 833, in read_array
    raise ValueError("Object arrays cannot be loaded when "
                     "allow_pickle=False")
ValueError: Object arrays cannot be loaded when allow_pickle=False

[gmgunter]

Looks like your formatter barfed here. Can we fit this on a single line for readability?

Maybe something like

    fill_value = float(data["fill_value"]) if data["fill_value"] is not None else None
    return RegularGridInterpolator(
        ...,
        fill_value=fill_value,
    )