sightloom-analysis
Zone dwell/occupancy analytics, pattern miners, and pluggable anomaly backends for SightLoom.
= "0.1"
Anomaly backends (AnomalyDetector)
| Backend | Type |
|---|---|
StatisticalAnomalyDetector |
z-score, MAD, CUSUM, day-of-week seasonality |
IsolationForestDetector |
pure-Rust isolation forest |
OcSvmDetector |
pure-Rust RBF one-class SVM (not libsvm SMO) |
GraphRelationalDetector |
multi-camera hops, rare transitions, rare pairs/routes |
| FAR calibration | calibrate_far_from_series → z-threshold at target FAR |
| Scoped baselines | ScopedBaselineStore (subject + camera) + detect_statistical_scoped |
| Quantum stub | QuantumStubDetector (host wire-up placeholder, not real quantum) |
Heavier graph ML / quantum solvers still plug in behind the same trait.
Pattern miners
Time-of-day, day-of-week, visit periodicity, dwell distribution, route sequences, co-occurrence, expected absence, group formation, event-before-event.