expand cucim.skimage.segmentation benchmarks#323
expand cucim.skimage.segmentation benchmarks#323grlee77 wants to merge 1 commit intorapidsai:branch-22.08from
Conversation
cucim.skimage.segmentation modulecucim.skimage.segmentation benchmarks
cucim.skimage.segmentation benchmarks| with open(fbase + ".md", "wt") as f: | ||
| f.write(all_results.to_markdown()) | ||
| try: | ||
| import tabulate |
There was a problem hiding this comment.
Should we add this dependency somewhere?
There was a problem hiding this comment.
Not sure where the best place would be? We could add a requirements-bench.txt or something, but we should not add either tabulate or pandas to the main package dependecies as they aren't used outside of the benchmarks. We don't distribute these benchmark files with the conda or PyPI packages.
There was a problem hiding this comment.
Yeah we could do that.
Another option would be to add it to this environment file. We can comment sections and specify these are benchmark requirements for clarity.
There was a problem hiding this comment.
We do have a requirements-test.txt within the python/cucim folder which has various test requirements that are not listed in that env.yml. I can just add a requirements-bench.txt to the changes in #324
There was a problem hiding this comment.
Maybe we should include these in the env.yml as well
This PR expands the approach of #322 to the remaining `cucim.skimage` benchmarks as well This PR is built on top of a few other open benchmarking PRs, so please review and merge those first: (#322, #323, #290) Authors: - Gregory Lee (https://github.com/grlee77) - https://github.com/aasthajh Approvers: - https://github.com/jakirkham URL: #324
I noticed that benchmark coverage for the segmentation functions was missing several functions. This PR adds them.
Unfortunately functions in the segmentation module take different mixtures of image and label image inputs so there are multiple different benchmarking classes required to set up appropriate inputs. The appropriate benchmarking class is chosen based on the function name.