Learned detectors for AIMsight: models trained on the reviewed labels to tell, at each measurement instant of a station, which families are present (`labeling/families.md`), and which sensors carry them. A pilot, not yet in use: the detectors in production are the rules of `aimsight-detectors`.
## The pilot
- **Families**: Visibility and Obstruction; the others stay rules for now.
- **Model**: multiple-instance learning, since a label names a station and a range, not its sensors, and each station has its own number of sensors: one small MLP shared by every sensor, attention pooling to a score per family for the station; the attention weights give the sensors.
- **Inputs**: each sensor against its own usual level (median and MAD over the 3 days before), on windows in hours (the step goes from 5 min to 4 h); the station context (share of high and rejected sensors, `sigma0`, references, camera pose, housing temperature, common mode by PCA); the image quality metrics when present, with a mask.
- **Training**: pre-training on the outputs of the current detectors over all the data, then fine-tuning on the reviewed labels of `labeling/auto/`.
- **Evaluation**: the measure of `detectors/bench/score.py` (found, correct, per family and per instant), against the rules on the same instants.
- **The criterion** (decided 2026-11-16): on the stations seen, the way the model will run (every station followed keeps being labelled in the app): trained on the start of every station, tested on its end (a cut in time, never at random). The model beats the rules by at least 5 points of found and of correct, in total, for Visibility and for Obstruction, and loses nowhere more than 2 points on a family of a station (the regression of `detectors/CLAUDE.md`). A new station stays under the rules until it has a few weeks of reviewed labels; the score with a station left out is followed at every step, as an indicator, not a gate.
- **Steps**: dataset, evaluator (reproducing the rules' figures first), gradient boosting baseline, the PyTorch model, pre-training compared with and without, decoding into episodes.
`dataset.py` (step 1, ~2 min): one `cache/<STATION>.npz` per labelled station, over its labelled range as the benchmark reads it: `xs` [instant, sensor, feature], `xc` [instant, feature], `y` [instant, family] with the conventions of `bench/score.py` (its labelled instants per family are the benchmark's, exactly). NaN where a value does not exist (a 1 h window at a 4 h step, no temperature, no image quality outside LENZSPITZE).
`evaluate.py rules` (step 2, ~20 min, once per change of the detectors): the rules' predictions per instant in `cache/rules/`, and the check that their counts are those of `bench/score.py`, station by station (exit 1 otherwise). `evaluate.py score`: a folder of `<STATION>.npz` with `yhat` [instant, family] against the labels, beside the rules, with the regressions of `detectors/CLAUDE.md` (more than 2 points of found or correct lost on a family of a station).
`m1.py` (step 3, a few minutes): gradient boosting on the station's features and the aggregates of its sensors', Visibility and Obstruction, leaving one station out; `--split-time`, the diagnosis on the last 30 % of every station, all stations seen in training. First results (2026-11-16, found / correct, total):
| | Visibility | Obstruction |
|---|---|---|
| rules, all stations | 47 / 77 % | 26 / 28 % |
| M1, station left out | 44 / 62 % | 9 / 7 % |
| M1, station left out, `--balance` | 46 / 70 % | 21 / 23 % |
The information is in the features: with every station seen, M1 beats the rules; moving to a new station is what fails (Obstruction is labelled on three stations only). Against the criterion, M1 passes the totals but not the regressions: on the last 30 %, Obstruction found at BREITHORN (90 → 57 %, 21 instants), LENZSPITZE (15 → 9 %, 33) and SNCF01 (100 → 56 %, 9 instants, where the rules flag 272 for 9 right); Visibility at LENZSPITZE (found 64 → 58 %, correct 98 → 81 %), SNCF01 and WILDSTRUBEL (correct 96 → 89 % and 97 → 93 %, with found 10 → 68 % and 86 → 99 %).
- Reads the measurements through `detectors/scripts/data.py`, the reading of the benchmark; the labels from `labeling/auto/`; the data from `data/` (read-only). Paths are passed as arguments.
- Never changes `aimsight-detectors` (neither `scripts/` nor `bench/`): the benchmark is the measure.
- Episodes in the format of the detectors' contract (`start`, `end`, `kind`, `label`, `text`, `ignore`, `sensors`), so that a model can later sit behind `detect(measurements)`.
## Rules
- Every time is in UTC. No rule specific to a station, a target or a date.