-
-
Notifications
You must be signed in to change notification settings - Fork 5.1k
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
BUG: Regression in 1.11.3 can still fail for optimize.least_squares
with method='trf'
, maybe due to 0 lower bounds?
#19351
Comments
I had a pretty good suspicion that the problem wasn't entirely fixed as you can see on the release discussion here: #19279 (comment), and obviously from the issue you linked, which we knew about before I did the To get these issues squashed more decisively I probably need some domain expertise to chime in. I think we usually ping @nmayorov for this corner of @mdhaber also has more knowledge of optimization algorithms than I do in general (pretty sure at least). Any thoughts on the sequence of patches below and the issues that remain open? Patch 1: #18896 New issues that remain and/or popped up as a result of merging those patches: |
Hmm thanks but unfortunately I don't know much about these algorithms, either. I would defer to @nmayorov, too. |
I think the wisest decision is to restore the code prior to #18896 The original complaint (about x not being exactly 0) I believe is not substantial or at least not critical. It is rather a small issue or a special property of the algorithm, certainly not a bug. |
Maybe the docs could briefly explain the special property so folks don't think it is a bug? |
I plan to create a PR with the revert + the test adjustment + small clarification of what to expect from the algorithms. |
… close to bounds Related issue -- scipygh-19351
… close to bounds Related issue -- scipygh-19351
… to bounds for method 'trf' Essentially reverts commit ba761d8 Closes scipy#19351
… to bounds for method 'trf' Essentially reverts commit ba761d8 Closes scipy#19351
Describe your issue.
Hi @tylerjereddy (thanks for all the maintenance work!)
This issue follows up on #19103, maybe also echoes #19309 (but the error message is not linked to division by zero).
Optimization with
optimize.least_squares
andmethod='trf'
fails since1.11.2
, and for some cases it was not fixed in1.11.3
(but some were). We noticed this from our CI test suite in SciKit-GStat (https://github.com/mmaelicke/scikit-gstat).I'm not sure what the origin is, maybe related to lower bounds equal to zero?
Reproducible example below.
Reproducing Code Example
Here's an example that:
Error message
As for previous issues on this topic, the error message is (in both 1.11.2 and 1.11.3):
SciPy/NumPy/Python version and system information
SciPy 1.11.1 to 1.11.3.
Thanks for looking into this!
The text was updated successfully, but these errors were encountered: