Closed bmwoodruff closed 3 months ago
(I'm going to have a conversation with myself to demonstrate what might happen when you ask for a review. Many times I'll just make change directly so we can pass this along to Numpy).
Remember to specify that you built numpy and the docs, and then ran the doctests after installing numpy.
spin build --clean
python -m pip install .
spin docs --clean
python tools/refguide_check.py --doctests
Everything built, so hooray! Remember to specific so I know it's working before I make changes.
Comments:
The changes all built. I've pushed them. Here's are two branches:
@charris, I'd love your feedback. Here's some questions I have:
ma.convolve
: https://github.com/numpy/numpy/compare/main...otieno-juma:numpy:ai-examples-ma-convolve . It's ready for your review. I wanted you to have a single function, and batched version, of what we can provide, side-by-side. We can run this AI tool across all of the Numpy code base. Some times AI will say a function has sufficient examples. Sometimes it returns utter garbage. Other times it finds nice extra examples. Often, it does a decent job of extending the existing examples.
svd
was useful, as well as repeated eigenvalues for eig
. AI suggested the examples, which were trash, but a bit of revision from a human and done.outer
with a negative may help someone better understand how outer works. That example is definitely aimed at a newbie. svdvals
third example does a great job at illustrating how the svdvals
are ordered from largest to smallest (with the diagonal matrix ordered small to large).I don't want to overload the PR list (we're around 200, and I know you guys want to keep it to 150). That's why batching may be better.
Thanks for being willing to be voluntold by @teoliphant to help us out here.
I'm closing this issue. It as designed as a practice run. I've gotten the needed feedback from the NumPy maintainers at the triage meetings.
Please review the examples in the following branch
Close when the following are complete: