Proof of concept use contourpy for contouring #5923
Conversation
isn't contourpy the same code mpl uses, so wont it be the same anyway? |
There are three different algorithms:
I don't think there's any major differences between them, but if we use the default new algorithm for |
|
Would we be able to ask mpl to use the new one via a kwarg when they depend on contourpy? |
|
Yes, I think so |
|
For reference, here's the Matplotlib issue and pull request: |
|
Happy to change how we do it for now and drop back to the new mpl behaviour in ~18 months when we can depend on it in mpl? |
setup.cfg
Outdated
| < 8000 /td> | @@ -55,6 +55,7 @@ image = | ||
| jpeg2000 = | |||
| glymur>=0.8.18,!=0.9.0,!=0.9.5 | |||
| map = | |||
| contourpy>=1 | |||
There was a problem hiding this comment.
This should now be contourpy>=1.0.1. I had an unnecessary numpy version pin in contourpy which might cause problems, the new release fixes this.
|
Contourpy dev here. Do you folks want mpl's |
|
👋 If mpl switches to the new algorithm then I think it has the same effect for us, when we depend on that version of mpl we can say they do the same thing, so I think that approach is fine with us :) |
|
If we plan to work around the main issue that we have with scikit-image, do we just wait for MPL to depend on contourpy and we close this PR? |
|
I'm not bothered whether this makes it into 4.0; the only good reason to rush seems to be removing the |
|
I am happy to get this into 4.0 if we decide that removing scikit-image is important enough, if not, waiting for MPL to do the switch is just easier for us? (We would still need to update the contour function tho) |
|
I think we should hold off on this until MPL makes the switch that won't be before we feature freeze 4.0, so I'm going to close this PR and we can track this at #5874 |
See #5874. It didn't take me long to throw this together so thought I'd open a draft PR for discussion. Things we should consider (this is non-exhaustive and off the top of my head; please add to this):
.draw_contour()to generate the contours throughcontourpy, to make them consistent with the results ofcontour()?