delays

Module for computing the detection delay in online change point detection.

Evaluates how quickly an online algorithm reacts to a true change point. Missed change points are penalized with the maximum allowable delay.

class pysatl_cpd.analysis.metrics.single_run.online.delays.Delays(max_delay)

Bases: ISingleRunMetric[TraceT, ProviderT, list[int]], Generic

Computes the detection delay for online change point detection algorithms.

Delay is defined as the distance between an actual change point and its corresponding detection. If a change point is missed (False Negative), the algorithm is penalized with the maximum defined delay.

Parameters:

max_delay (int) – The maximum allowable delay window for a valid detection. Also used as the penalty value for missed detections (FN).

Raises:

ValueError – If maximum delay is negative

__init__(max_delay)

Parameters:

max_delay (int)

Return type:

None

evaluate(run)

Calculate delays for all true change points

Delay is computed per true change point as the minimum non-negative delay among matched detections within [true_change, true_change + max_delay]. If there is no match, max_delay is returned for that change point.

Parameters:

run (SingleRun[TypeVar(TraceT, bound= OnlineDetectionTrace), TypeVar(ProviderT, bound= LabeledData)]) – The run to evaluate.

Returns:

A list of delays where each element corresponds to the true change point at the same index in data.change_points. Length is exactly equal to the number of true change points.

Return type:

list[int]