When I've been reading articles about change points there's a lot of discussions about performance. But I don't see a lot of benchmarks here in ruptures.
I suggest to use airspeed velocity (used by numpy/pandas/xarray etc) and make some simple tests that loops through each of the algorithms.
asv can be difficult to setup locally (at least for me) so I suggest to add it to the CI as well to avoid complexity for new (and old) contributors:
xarray has a nice workflow that you can use (PR: https://github.com/pydata/xarray/pull/5796) and I suggest to try to follow the same folder structure as xarray does.
deepcharles
commented
2 years ago
When I've been reading articles about change points there's a lot of discussions about performance. But I don't see a lot of benchmarks here in
ruptures.I suggest to use airspeed velocity (used by numpy/pandas/xarray etc) and make some simple tests that loops through each of the algorithms.
asv can be difficult to setup locally (at least for me) so I suggest to add it to the CI as well to avoid complexity for new (and old) contributors: xarray has a nice workflow that you can use (PR: https://github.com/pydata/xarray/pull/5796) and I suggest to try to follow the same folder structure as xarray does.
Interesting reading: https://labs.quansight.org/blog/2021/08/github-actions-benchmarks/