UC Berkeley (BIDS) sprint, May 28 Jun 2 2018
This page collects ideas and issues for the scikit-image side of the joint scikit-learn/scikit-image/dask sprint at UC Berkeley.
- work on the parallelization of some algorithms to improve performance, focussing maybe first on the case of machines with several cores (rather than a distributed architecture, which most users don't use I guess). This would need some benchmarking and we could test several solutions such as dask (with Matt Rocklin, we could test and maybe improve our apply_parallel) as well as joblib and its different backends (with Gaël and Olivier). Emma can provide some tomography datasets for this, it would be better if we had a machine with 10-20 cores to ssh on for benchmarking, maybe with AWS? Or a BIDS machine?
- discussing all GitHub "needs decision" PRs/issues.
- get nD transforms, based on NumPy array coordinates, working. (with Kira Evans)
- figure out whether we can reliably distribute Numba, and if so, do we want to get that dependency.
- implement flood-fill with support for lowlevelcallables (see https://github.com/scikit-image/scikit-image/issues/2876#issuecomment-385241507)
- fix data types and ranges. (See #3009.)
- talk about release schedules, governance, funding, leadership, outreach, and a few more general topics. (Though Juan thinks these things should happen in the evenings over dinner etc. ;)
Stuff that we want to happen before the sprint:
- announce the sprint on the mailing list to see whether anyone wants to join remotely
- get airspeed velocity working (though maybe this needs to happen before the sprint anyway)
- draw up scikit-image 1.0 roadmap (including emailing mailing list for community discussion)