ENH: Add full_output support to optimize.newton() #8803
Conversation
Added keyword args full_output=False, disp=True to optimize.newton(...) to better match the other zero finders. The default values preserve the current behavior. Setting full_output=True causes a RootResults object to be returned with information on the total number of function_calls and iterations. Setting disp=False suppresses the RuntimeError associated with non-convergence, and is likely best combined with full_output=True.
| # Newton-Raphson method | ||
| def newton(func, x0, fprime=None, args=(), tol=1.48e-8, maxiter=50, | ||
| fprime2=None): | ||
| fprime2=None, | ||
| full_output=False, disp=True): |
There was a problem hiding this comment.
New keyword arguments need to be added last (otherwise they can break backward compat)
There was a problem hiding this comment.
Aren't the new keywords full_output and disp added last? fprime2 was the last argument, and full_output and disp appear after fprime2.
There was a problem hiding this comment.
Ok, sorry I didn't notice also disp was new...
There was a problem hiding this comment.
No problem, thanks for looking at it.
| msg = "Failed to converge after %d iterations, value is %s" % (maxiter, p) | ||
| raise RuntimeError(msg) | ||
| funcalls += 1 | ||
| if disp: |
There was a problem hiding this comment.
... this then becomes if not full_output:.
| True, the return value is ``(x, r)``, where `x` is the root, and `r` is | ||
| a RootResults object. | ||
| disp : bool, optional | ||
| If True, display a RuntimeError if the algorithm didn't converge. |
There was a problem hiding this comment.
Not sure if disp is the right name for this, because the old behavior is to raise an exception on convergence failure, not only display it. Typically I think disp was used to print progress during iterations to stdout, which is not what it means here.
I'd suggest removing the disp keyword argument, and retain the error-raising behavior only for not full_output. This is how it works e.g. in fsolve and elsewhere, so here we probably should follow suit.
There was a problem hiding this comment.
One reason I added the disp keyword was for compatibility with all the other solvers in zeros.py: bisect, brentq, brenth, ridder. It is also the case that testing all root-finder methods within a uniform testing framework is much simpler if both the full_output and disp keywords are present.
I looked for the actual usages of disp and found call_solver in optimize/zeros.c which sets PyExc_RuntimeError with an error message and returns NULL if the solver fails to converge. I didn't find any progress reporting in these solvers, only reporting of convergence errors.
Are you suggesting that perhaps all the zeros solvers (brentq/ridder/...) should match fsolve and not have the disp keyword?
There was a problem hiding this comment.
Ah, I did not remember the 1D solvers are inconsistent with fsolve. Then I guess it's best to have this as it is and follow the 1D solver example.
There was a problem hiding this comment.
Tests for full_output and disp options are now added; newton() consistent with the 1D solvers; pytest.raises is used instead of numpy.testing.assert_raises_regex. Any other changes needed?
scipy/optimize/tests/test_zeros.py
Outdated
| from numpy.testing import assert_raises | ||
|
|
||
| def _assert_raises_regex(exception_class, _, *args, **kwargs): | ||
| assert_raises(exception_class, *args, **kwargs) |
There was a problem hiding this comment.
You can use with pytest.raises(SomeException, match="message regex"): some_call(...) instead.
Numpy's assert_raises requires nose, which we'd like to deprecate eventually.
Use pytest.raises() instead of numpy._assert_raises_regex() Replace transcendental functions with polynomials when testing the number of iterations in optimize.zeros.newton() to avoid platform-dependent exp/sin/cos.
|
Thanks, merged. Sorry for not fully awake review... |
Signed-off-by: Mark Mikofski <[email protected]>
Added keyword args
full_output=False,disp=Truetooptimize.newton(...)to better match the other zero finders. The signature is now
The default values preserve the current behavior.
Setting
full_output=Truecauses aRootResultsobject to be returned withinformation on the total number of function_calls and iterations.
Setting
disp=Falsesuppresses the RuntimeError associated withnon-convergence, and is likely best combined with
full_output=True.