Best so far swap - BestSoFarRootFinder should not be recommended to wrap LeastSquaresSolver

[johannahaffner]

This fixes a typo in the documentation. BestSoFarLeastSquares was undocumented, instead BestSoFarRootFinder was the wrapper listed in the section on least squares solvers.

I know that I used it to wrap LevenbergMarquardt, and I just checked - this will result in an unexpected solver instance.

import optimistix as optx


solver = optx.GaussNewton(rtol=1e-3, atol=1e-06)
wrapped_ls = optx.BestSoFarLeastSquares(solver)
wrapped_rf = optx.BestSoFarRootFinder(solver)

print(isinstance(wrapped_ls, optx.AbstractLeastSquaresSolver))  # True
print(isinstance(wrapped_rf, optx.AbstractRootFinder))  # True
print(isinstance(wrapped_rf, optx.AbstractLeastSquaresSolver))  # False 

In this case, it does not appear to do anything as long as optx.least_squares is used as the top-level API, because this will only check for a minimiser.

However, if optx.root_find is the top-level API and a wrapped least-squares solver is used, then the conversion to optx.least_squares won't happen, because this line will not evaluate to True:

optimistix/optimistix/_root_find.py

Line 197 in 1d8e79b

elif isinstance(solver, AbstractLeastSquaresSolver):

[johannahaffner]

I'll rebase on dev later, sorry for the oversight!

[patrick-kidger]

LGTM!

For this one you can merge straight to main.

The reason for a dev branch is that any changes to the docs on main are automatically reflected in the online documentation: so if we introduce a new feature then it should be on a dev branch to be sure that it is not advertised until it is released.

However conversely, doc fixes can/should go straight to main as there is no reason to delay fixing typos!

(So I think I'm still seeing the verbose changes on this branch. If you can make this branch have just the commit with the fix then I'll merge it immediately :) )

[johannahaffner]

(So I think I'm still seeing the verbose changes on this branch. If you can make this branch have just the commit with the fix then I'll merge it immediately :) )

That is why I rebased on dev. I'll figure out how to make my main forget it has the verbose changes :)

However conversely, doc fixes can/should go straight to main as there is no reason to delay fixing typos!

Makes sense!

[johannahaffner]

I also removed three lines in optx.root_find that seemed to be holdovers and occurred after a return statement in the elif branch.

[patrick-kidger]

Ah, good catch! Actually, those three lines should be there and it's the return statement that's wrong. I've just opened #99 to correct this. (And I've included your BestSoFarLeastSquares fix whilst I'm at it!)

[johannahaffner]

Ah, I did not catch this! Good that you did :) I'll close this PR.