maxime.pasquier pushed to main at maxime.pasquier/aimsight-data 2026-11-19 11:55:33 +01:00
73567fe585 stations.txt: FLORIDA at Ernen, for its weather (MeteoSwiss BIN Binn, 3.6 km)
799a53c102 pumping.py: start/running/log with diff, pump_diff.sh into data_diff/ (its own lock and log), for the Stations page of aimsight-app
ca366c150e Runs in data_diff/<SERIAL>/runs/<run>/: new_run.sh creates one, run_config.sh <SERIAL> <run> computes it
d3bf4b6860 Configurations: processing/run_config.sh recomputes configs/<name>.yaml from data_diff/ into runs/<name>/, read like data/
81908ca370 Pumping of the differential images: extraction/pump_diff.sh into data_diff/<SERIAL>/, in the tree of data/ (never data/ itself)
Compare 11 commits »
maxime.pasquier pushed to main at maxime.pasquier/aimsight-agents 2026-11-18 17:57:06 +01:00
59da8532ab analyst, support: the cases (labeling/auto/cases/, scripts/cases.py) replace the catalogue of phenomena and the false friends
4cd56ac172 analyst: the family Perturbation
e35d9650eb analyst: a missing point is not an anomaly (the station filtered it), a well compensated thermal movement is not labelled; corr_x_mm, corr_y_mm are the same curves in mm, same rule and threshold; few labels, each significant
de53bf1a56 analyst: an anomaly is what stands out from the noise of the corr (the threshold of labeling-rules.md, section 1); section 2 of the analysis gives the deviation in noises of what is left without a label
7aaabfdca0 support: a generalist agent that does what it is asked, delegating a range's reading to the analyst
Compare 5 commits »
maxime.pasquier pushed to main at maxime.pasquier/aimsight-tools 2026-11-18 17:56:46 +01:00
9560502dc5 GUIDE: the false friends and the phenomena are the cases (labeling/scripts/cases.py --list)
a9948d3404 improve: S1
fc305697f7 improve: S1
e03d911433 improve: S1
f57e6ea045 improve: S2
Compare 21 commits »
maxime.pasquier pushed to main at maxime.pasquier/aimsight-libredesk 2026-11-18 16:35:58 +01:00
a836fb4eac commande docker
maxime.pasquier pushed to main at maxime.pasquier/aimsight-libredesk 2026-11-18 15:40:59 +01:00
01317cb2f8 Conversations : tri conversations.updated_at (en cours, le pompage réel ne marche pas encore)
2dc3442cc4 Conversations : la clé de lecture sous LIBREDESK_API_KEY / LIBREDESK_API_SECRET, les noms du .env
e3c4a29e90 Fusion de embeddings-token dans main (le jeton devant Ollama, avec le pompage des conversations)
0fa7c4936b Image embeddings-bge-m3 : jeton exigé devant Ollama
Compare 12 commits »
maxime.pasquier pushed to main at maxime.pasquier/aimsight-detectors 2026-11-18 10:56:33 +01:00
1ca6557af2 displacement: the text of a jump back says nothing to do (its family, Perturbation); the docstring notes that the return is read on the corrected curves only, so a camera that stays while the curves come back is a Perturbation here, a Displacement in the labels (for the optimizer); figures unchanged
47bc0acde2 Perturbation, the seventh family: a jump back to normal is a Perturbation (nothing to do, greyed), Displacement keeps the lasting shift, the jump not back yet and the slow drift (labeling spec 2026-11-18-displacement-split-design); benchmark: 0 regression, Perturbation found 0 -> 4 %, +1 code line
428ebaacfe categories.window: the events of a range with its history, for aimsight-app and the CLI (figures unchanged, +3 code lines)
a008ea8fcc Regression computed by bench/compare.py (Total > 1 point; station > 5 points and > 20 instants), complexity in the benchmark, pruning after a fix
63d954a126 episodes.py, not common.py: the tools repository has its own common.py on the same sys.path in the app
Compare 18 commits »
maxime.pasquier pushed to main at maxime.pasquier/aimsight-labeling 2026-11-18 10:51:39 +01:00
8e4ebca342 families.md at HEAD: the phenomena that come back move under Perturbation, the camera step stays a Displacement when it stays (the cases session's pending commit removes this section for the cases; until then HEAD no longer contradicts itself)
de2bc4a5ab auto: CLARIDEN-12 no longer points to a thermal label that does not exist (review removed); only this line, the rest of CLARIDEN's pending re-analysis left as it is — completes de092c8, whose message already named it
de092c8008 auto: CLARIDEN-12 and MICHIGAN-19 no longer point to labels that do not exist, WILDSTRUBEL-136 no longer repeats its return (review removed on the reworded CLARIDEN-12 and MICHIGAN-19); the analysis copies of the reviews aimsight-app wrote in the 9 stations committed in 9ccd04d, and the removal of WILDSTRUBEL-70's copy, so that each copy matches its final label at HEAD
fed05a7d10 Fixes of the final review: families.md gives one answer for an unexplained movement that returns (Perturbation); NORDEND-257 keeps its hypotheses next to "not established"
Compare 19 commits »
maxime.pasquier pushed to main at maxime.pasquier/aimsight-models 2026-11-17 11:59:50 +01:00
4836929bc7 Steps 5 and 6: pre-trained M2 passes the criterion for Obstruction, not for Visibility; episodes and sensors (decode.py)
maxime.pasquier pushed to main at maxime.pasquier/aimsight-models 2026-11-17 09:51:57 +01:00
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
maxime.pasquier renamed repository from Support-LibreDesk to maxime.pasquier/aimsight-libredesk 2026-11-17 09:38:10 +01:00
maxime.pasquier pushed to main at maxime.pasquier/aimsight-models 2026-11-16 20:54:48 +01:00
7312a1c3ca M2 results: no better than M1 on stations seen, worse on a station left out
maxime.pasquier pushed to main at maxime.pasquier/aimsight-models 2026-11-16 19:22:54 +01:00
fef30b5ef4 Step 4: M2, multiple-instance model in PyTorch (shared sensor MLP, attention per family, averaged seeds); the time-split report shared with M1
maxime.pasquier pushed to main at maxime.pasquier/aimsight-models 2026-11-16 17:59:08 +01:00
975ebc7c1a The criterion: stations seen, cut in time; M1's regressions station by station; no regression when nothing is flagged
maxime.pasquier pushed to main at maxime.pasquier/aimsight-models 2026-11-16 17:48:13 +01:00
f93cdf3cd1 Step 3: M1, gradient boosting on station aggregates (leave one station out, and a time split)
maxime.pasquier pushed to main at maxime.pasquier/aimsight-models 2026-11-16 16:25:05 +01:00
c04cbc90bb Step 2: the evaluator (the measure of bench/score.py on predictions per instant; the rules reproduce it exactly)
maxime.pasquier pushed to main at maxime.pasquier/aimsight-models 2026-11-16 15:49:06 +01:00
b2d3734152 README: how to build the dataset
maxime.pasquier pushed to main at maxime.pasquier/aimsight-models 2026-11-16 15:48:53 +01:00
b6eecf3ab7 Step 1: the dataset (features per sensor and per station, families per instant as the benchmark)
maxime.pasquier pushed to main at maxime.pasquier/aimsight-app 2026-11-16 15:43:29 +01:00
8fa555f43b The models/ link: the pilot of learned detectors (aimsight-models)
maxime.pasquier pushed to main at maxime.pasquier/aimsight-models 2026-11-16 15:40:31 +01:00
9cf20766a5 aimsight-models: the pilot of learned detectors (README)
maxime.pasquier created branch main in maxime.pasquier/aimsight-models 2026-11-16 15:40:31 +01:00