Change default for window size in EquivalentSourcesGB

[indiauppal]

Change the default value for the window_size in the EquivalentSourcesGB constructor to "default": window size will be estimated to locate approximately 5000 data points in each window. If window_size="default" and less than 5000 data points are used to fit the sources, a single window for all data and sources will be used. Check if the passed value of window_size is valid. Add a new window_size_ attribute to the class that is set after fitting the sources. Update docstrings and extend tests.

Relevant issues/PRs:
Fixes #425

[indiauppal]

• Add a test to check window size default. Use numpy.testing.assert_allclose

[indiauppal]

• Raise an warning for no. of points < 5e3

Completed with @santisoler

[santisoler]

This is starting to look great @indiauppal!

I'm leaving a few ideas after the meeting we had today:

1. We should probably extend the test function for less than 5000 points and test if the outputs match the ones for a single window. Something like:

   data_windows, source_windows = eqs._create_windows(grid_coords)
   # The output lists should have a single element each (corresponding to the single window)
   assert len(data_windows) == 1
   assert len(source_windows) == 1
   # Check if all sources and data points are inside the window
   for coord in grid_coords:
       npt.assert_allclose(coord, coord[data_windows[0]])
   for coord in eqs.points_:
       npt.assert_allclose(coord, coord[source_windows[0]])

2. We could explore replacing those np.aranges with slices. We only need those indices to slice the sources and coordinates arrays in the self._gradient_boosting method, so anything we could use to slice them that are more memory efficient should work. This is just a minor optimization, so it's not that critical for this PR.
3. Since now we have another attribute the fit method assigns (the window_size_), we should add it to the list of Attributes in the class docstring. Something like the think I wrote in Change default value of depth in equivalent sources #491 for the depth_ argument:
   

   harmonica/harmonica/_equivalent_sources/cartesian.py

   Lines 142 to 145 in 55309c7

   	depth_ : float or None
   	Estimated depth of the sources calculated as 4.5 times the median
   	distance between first neighboring sources. This attribute is set to
   	None if ``points`` is passed.

4. It would be nice to add a check for the window_size argument in the __init__ method. Basically, we should raise an error if the passed value is a str and is not "default". That's the only string value that should be valid. Check this out for inspiration:

   harmonica/harmonica/_equivalent_sources/cartesian.py

   Lines 176 to 180 in 55309c7

   	if isinstance(depth, str) and depth != "default":
   	raise ValueError(
   	f"Found invalid 'depth' value equal to '{depth}'."
   	"It should be 'default' or a numeric value."
   	)

This is somewhat my personal wishlist for this PR, so feel free to assign me a few of these tasks if you want. As always, feel free to ask for help if you need it 🙂

Looking forward to see this merged!

[santisoler]

@indiauppal, I'm updating this branch after the fix I made for the failing Mac testing. Remember to run a git pull to sync your local repo with the latest change in this branch.