-
Notifications
You must be signed in to change notification settings - Fork 0
Open
Description
Link score-based and rank-based:
- implement exhaustive score to partial order count
- implement MonteCarlo score to partial order count
- Find and implement adaptive / online technique
Plackett-Luce Sinkhorn algorithms :
- implement sinkhorn problem builder
- implement sinkhorn problem solver (currently using pot solvers)
- implement sinkhorn log problem builder
- All of the above => Plackett-Luce order
Condorcet compliant orders :
- implement pair order count to condorcet order (until does not exist)
- Implement Kemeny-Young order ? https://en.wikipedia.org/wiki/Kemeny_method
- Implement ranked pairs order https://en.wikipedia.org/wiki/Ranked_pairs
- Implement Schulze order https://fr.wikipedia.org/wiki/M%C3%A9thode_Condorcet_avec_rangement_des_paires_par_ordre_d%C3%A9croissant
Other orders :
- Just sort it! cf https://arxiv.org/abs/1502.05556
Ranking benchmarks :
- Figure out how to store ranking benchmarks
- Figure out how to reduce ranking benchmarks (most timings / score-based reductions do not make sense)
- Figure out the best way to plot ranking benchmarks
Manual rank assign:
- compare_intent values "group" and "compare"
- compare_intent value "restricted" vs "unrestricted"
Image ranking app:
- gather 2 - way best count
- gather k - way best count
- gather k - way order count
Build, test and docs :
- Figure out local cython builds
- Figure out cibuildwheels for different platforms on Gihub Actions
- Figure out conda recipe changes and distribution subtleties
- Clean API
- Tests
- Docs
- Create changelog
Metadata
Metadata
Assignees
Labels
No labels