train_all() now raises RainUnavailableError when use_rain=True and the
Open-Meteo history cannot be loaded, instead of logging a warning and
writing gauge-only (v2) bundles over the deployed v3 set -- which is what
the 2026-09-01 server retrain did unnoticed. --no-rain remains the explicit
way to get v2. CLI exits 2 with a one-line error. Three tests cover the
guard, the opt-out, and the v3 happy path.
scripts/retrain.sh trains into models/.staging, refuses to promote unless
metrics.json shows hgb-v3+ and >=14 trained stations, then renames bundles
into place (previous generation kept in models/.previous). No API restart:
predict.py reloads by mtime on the hourly precompute.
water-monitor-retrain.{service,timer}: 1st of each month 03:30, Persistent,
OMP_NUM_THREADS=4, Nice=15, same sandbox as the API unit. install.sh now
does `uv sync` into .venv (one env rule; removes a stale venv/) and enables
the timer. water-monitor.service in the repo matched neither the deployed
unit nor the uv env; it now does (run.py --web-api, .venv, EnvironmentFile).
Add src/ml/ package predicting, per station and per 6/12/24 h horizon,
the probability of exceeding warning (3.0 m) and danger (4.5 m) levels
plus expected peak level, trained on the 592k-row PostgreSQL history:
- features.py: hourly grid with coverage gating and no future leakage;
upstream stations enter at empirically measured travel-time lags
(P.20 +17h ... P.103 +1h vs P.1); hour-of-day deliberately excluded
(it encodes the scrape schedule, not hydrology)
- train.py: HistGradientBoosting regression + warn/danger classifier
heads per station x horizon, >=30-positives gate with calibrated
sigmoid-on-regression fallback, strict temporal splits, per-event
lead-time evaluation; guards against sklearn 1.9.0 crash on
degenerate feature columns
- predict.py: bundle loading with feature-name checks, heuristic
fallback tier, get_latest_forecasts() for the API; raises when no
models are trained so the endpoint 503s instead of serving
persistence output as forecasts
- data.py: Postgres-first loader (FLOOD_ML_DB_URL override), HTTP API
fallback (flagged: that path backfills synthetic discharge), csv.gz
cache
- /forecast endpoint (15-min TTL cache) + dashboard flood-risk panel
(hidden until models exist)
- docs/FLOOD_FORECASTING.md: full system doc with measured deployment
numbers (~335 MB RSS, CPU negligible, ~6 min full retrain) and
retraining policy
Validation: out-of-sample backtest of the record 2024 flood season
(train <= Aug 2024) alerted 24-48 h ahead of the Oct 5 peak; 2025-26
test split: P.1 6h PR-AUC 0.974, recall 98.3% at 1% false-alarm rate.
Also: fix P.81 station coordinates (was Ban Pong/Ratchaburi, 493 km
out of basin; now 18.6936 N 99.0819 E per RID station page), pin
scikit-learn==1.9.0 and numpy<2, gitignore model artifacts (~100 MB,
train on the server via scripts/train_flood_model.py).
grabowski