state_arl

ARL-by-state scenario for no-reset benchmarks.

class pysatl_cpd.benchmark.online.noreset.scenarios.state_arl.NoResetArlByStateScenario(entries, collect_states=False, state=None, arl_length=0)[source]

Bases: NoResetBenchmarkScenario[DataT, DataFrame]

Scenario that evaluates ARL for a specific state across thresholds.

Parameters:

  • entries (Sequence[OnlineNoResetBenchmarkEntry]) – Detector entries to benchmark.

  • collect_states (bool) – Whether to retain algorithm states during detection (default False).

  • state (StateDescriptor | None) – Target state for no-change providers.

  • arl_length (int) – Expected length of each no-change run. Must be positive.

Raises:

ValueError – If state is None or arl_length is not positive.

__init__(entries, collect_states=False, state=None, arl_length=0)[source]
Parameters:
Return type:

None

prepare_benchmark_jobs(dataset)[source]

Build ARL benchmark jobs for each entry.

Filters the dataset to segments matching the target state, creates a no-change provider dataset of the requested length, and builds one job per entry.

Parameters:

dataset (Dataset[TypeVar(DataT), TimeseriesAnnotation]) – Input dataset with segment annotations.

Returns:

One job per entry, each using the no-change providers.

Return type:

Sequence[BenchmarkJob[TypeVar(DataT)]]

Raises:

ValueError – If no segments match the target state.

set_registry(registry)[source]

Set the registry used by the internal ARL analyzer.

Return type:

None

Parameters:

registry (BenchmarkRegistry[DataT, OnlineDetectionTrace[Any]])

set_classification_report(classification_report)[source]

No-op; ARL scenario does not require a classification report.

Return type:

None

Parameters:

classification_report (NoResetClassificationReport[Any, Any])

analyze(registry)[source]

Evaluate ARL for every entry using resolved thresholds.

Parameters:

registry (BenchmarkRegistry[TypeVar(DataT), OnlineDetectionTrace[Any]]) – Registry containing cached detection runs.

Returns:

ARL table per detector description, or empty dict if no ARL providers were registered.

Return type:

dict[ChangePointDetectorDescription, DataFrame]

property state_checked: StateDescriptor

Validated state descriptor; raises if not set.