Skip to content

Calibration CI: groups that fit, and a timeout for each - #987

Merged
derrynknife merged 1 commit into
devfrom
fix/calibration-ci-time
Oct 17, 2026
Merged

derrynknife merged 1 commit into
devfrom
fix/calibration-ci-time

Conversation

@derrynknife

Copy link
Copy Markdown
Owner

The release run's calibration jobs (#984) ran for hours.

  • conformance-sweeps. The cause was ExpoWeibull's mrl_cb[lr] sweep. Its mean residual life is an integral at every point the search tries, in a three-parameter valley. One two-sided bound at one time took 414 s locally on four cores, and the sweep asks for dozens of them.
    • It is now left out of the sweep, as the offset, LFP and ZI variants' mrl_cb[lr] already were for the same reason.
    • mrl_cb[lr] is still swept on 36 cases, and the Weibull's takes the same path in closed form.
    • The rest of the sweep: 242 tests, 10.4 min on four workers locally.
    • The joint model is not in this sweep.
  • extra-parameters. Its four studies took 52 min in one group on the first release run: offset 1284 s, on the edge 903 s, lfp 651 s, zi 321 s. They now run as three groups (offset; on the edge; lfp + zi), each about 21, 15 and 16 min at those timings. No repetitions are cut.
  • Timeout. Every calibration group gets timeout-minutes: 45, so a hang fails fast instead of running for GitHub's six hours.

Separately: a 414 s ExpoWeibull mrl_cb(method="lr") is slow for users too. That is noted as follow-up, not changed here.

Local: fast conformance passes; the sweep passes (242) in 10.4 min.

🤖 Generated with Claude Code

https://claude.ai/code/session_019Km2eWBiPJ5JNkme3vRg4B

The release run's calibration jobs ran for hours:

- conformance-sweeps: the ExpoWeibull's mrl_cb likelihood-ratio sweep.
  Its mean residual life is an integral at every point the search
  tries, in a three-parameter valley; one two-sided bound at one time
  took 414 s on four cores, and the sweep asks for dozens. It is left
  out, as the offset, LFP and ZI variants' already were for the same
  reason; mrl_cb[lr] is still swept on 36 cases, the Weibull's taking
  the same path in closed form. The rest of the sweep: 242 tests, 10.4
  minutes on four workers locally.
- extra-parameters: its four studies took 52 minutes in one group
  (offset 1284 s, on the edge 903, lfp 651, zi 321). They run in three
  groups now, offset, on the edge, and lfp with zi, none of them cut.

Every calibration group gets timeout-minutes: 45, so a hang fails in
three quarters of an hour rather than GitHub's six.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019Km2eWBiPJ5JNkme3vRg4B
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.

1 participant