MultiSensorValidationEvents¤

Cross-checks redundant sensors that measure the same variable: windows where they disagree, pairwise bias, a consensus score, and which sensor deviates from the group.

ts_shape.events.quality.multi_sensor_validation · Quality guide · API reference

Example¤

import ts_shape
from ts_shape.events.quality.multi_sensor_validation import MultiSensorValidationEvents

sensors = ["temp_1", "temp_2", "temp_3"]
df = ts_shape.make_timeseries(sensors, n_points=1440, freq="1min", noise=0.1, seed=None)
df.loc[df["uuid"] == "temp_3", "value_double"] += 0.8  # temp_3 reads high

msv = MultiSensorValidationEvents(df, signal_uuids=sensors)
disagreements = msv.detect_disagreement(threshold=0.5, window="5min")
bias = msv.pairwise_bias(window="1h")
print(msv.detect_outlier_sensor(window="1h", method="median").head())

Methods¤

Method	Returns (per window)
`detect_disagreement(threshold, window='5min')`	windows where max − min across sensors exceeds `threshold`: `max_spread`, `sensor_high`, `sensor_low`, `duration`
`pairwise_bias(window='1h')`	one row per sensor pair: `sensor_a`, `sensor_b`, `bias`, `abs_bias`
`consensus_score(window='1h')`	`consensus_mean`, `spread_std`, `consensus_score` (1.0 = perfect agreement, 0.0 = high disagreement)
`detect_outlier_sensor(window='1h', method='median')`	`outlier_sensor`, `deviation`, `other_sensors_mean`

Notes¤

Telling which sensor is off needs at least three sensors; windows with fewer reporting are skipped with a DataQualityWarning. On ties the first sensor in signal_uuids wins.
method is "median" or "mean" (the group center).
Take the detect_disagreement threshold from the largest spread seen in a known-good period plus a margin.