4 Commits
Author SHA1 Message Date
maximeandClaude Opus 5.5 5337dd0dac Step 5: pre-training on the rules' outputs (pretrain.py, m2.py --pretrain); regressions as detectors/bench/compare.py; the reader from the tools repository
The detectors were simplified on 2026-11-16: the reader moved to aimsight-tools (scripts/data.py, put on the path by
categories.py) and a regression is now a Total losing more than 1 point, or a family of a station losing more than
5 points and more than 20 instants.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-17 09:51:56 +01:00
maximeandClaude Opus 5.5 fef30b5ef4 Step 4: M2, multiple-instance model in PyTorch (shared sensor MLP, attention per family, averaged seeds); the time-split report shared with M1
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 19:22:53 +01:00
maximeandClaude Opus 5.5 975ebc7c1a The criterion: stations seen, cut in time; M1's regressions station by station; no regression when nothing is flagged
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 17:59:06 +01:00
maximeandClaude Opus 5.5 c04cbc90bb Step 2: the evaluator (the measure of bench/score.py on predictions per instant; the rules reproduce it exactly)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
2026-11-16 16:25:03 +01:00