Skip to content

Update FavMac - #1219

Open
lehendo wants to merge 1 commit into
sunlabuiuc:masterfrom
lehendo:favmacaudit
Open

Update FavMac#1219
lehendo wants to merge 1 commit into
sunlabuiuc:masterfrom
lehendo:favmacaudit

Conversation

@lehendo

@lehendo lehendo commented Aug 30, 2026

Copy link
Copy Markdown
Collaborator

Added:

  • The exchangeability/no-distribution-shift caveat to the public FavMac docstring, consistent with CovariateLabel/NeighborhoodLabel/SCRIB.
  • Input validation for target_cost > 0, delta in (0,1), and target_cost <= C_max (per the paper's c in (0, C_max] requirement)

…ation

Independently re-verified FavMac's two threshold formulas directly
against Lin, Trivedi, Xiao, and Sun, "Fast Online Value-Maximizing
Prediction Sets with Conformal Cost Control" (ICML 2023,
arXiv:2302.00839):

- Expected cost control: derived target_cost*(n+1) - C_max fresh from
  the paper's Eq. 18, confirmed exact. This branch adds the missing
  citation/explanation for this branch (favscrib only documented the
  violation-control branch's "-1" term).
- Violation control: re-derived the "-1" term fresh from Eq. 21,
  independently confirming favscrib's existing resolution is correct.

No new bug found in the core algorithm. Added:
- The exchangeability/no-distribution-shift caveat to the public FavMac
  docstring, consistent with CovariateLabel/NeighborhoodLabel/SCRIB.
- Input validation for target_cost > 0, delta in (0,1), and
  target_cost <= C_max (per the paper's c in (0, C_max] requirement) --
  previously unchecked, so invalid inputs would silently produce
  nonsensical thresholds instead of a clear error.

Verified: the internal engine's docstring doctest passes for real
(13/13), the expected-cost formula matches a direct re-derivation
exactly, and the new validation correctly accepts/rejects inputs.

Deliberately scoped to avoid touching the same lines already fixed on
favscrib, so the two branches merge independently.
@fbonc

fbonc commented Sep 1, 2026

Copy link
Copy Markdown
Contributor

looks good.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants