Evaluation

The tsseg.metrics namespace provides classes implementing common segmentation quality measures. Every metric inherits from tsseg.metrics.BaseMetric and shares the same compute(y_true, y_pred) contract (with an update / compute variant for streaming use).

Change-point metrics

These metrics compare two sets of sparse change-point indices.

`tsseg.metrics.F1Score`	Computes the F1-score for change point detection.
`tsseg.metrics.Covering`	Computes the Covering score for a segmentation.
`tsseg.metrics.HausdorffDistance`	Computes the Hausdorff distance between two sets of change points.
`tsseg.metrics.GaussianF1Score`	Gaussian-weighted alternative to the classic F1 score.
`tsseg.metrics.BidirectionalCovering`	Bidirectional extension of the classical Covering metric.

State-labelling metrics

These metrics compare two dense label arrays of equal length, aligning predicted to ground-truth states when needed (e.g. Hungarian matching).

`tsseg.metrics.AdjustedRandIndex`	Computes the Adjusted Rand Index (ARI).
`tsseg.metrics.AdjustedMutualInformation`	Computes the Adjusted Mutual Information (AMI).
`tsseg.metrics.NormalizedMutualInformation`	Computes the Normalized Mutual Information (NMI).
`tsseg.metrics.WeightedAdjustedRandIndex`	Computes the Weighted Adjusted Rand Index (WARI).
`tsseg.metrics.WeightedNormalizedMutualInformation`	Computes the Weighted Normalized Mutual Information (WNMI).
`tsseg.metrics.StateMatchingScore`	Computes the State Matching Score (SMS).

Base protocol

class tsseg.metrics.BaseMetric(**kwargs)[source]

Bases: ABC

Base class for all metrics.

abstractmethod compute(y_true, y_pred, **kwargs)[source]

Computes the value of the metric.

Parameters:

y_true (ndarray) – Ground truth labels or change points.
y_pred (ndarray) – Predicted labels or change points.
**kwargs – Additional arguments for metric computation.

Return type:

dict[str, float]

Returns:

A dictionary containing metric names and their values.

Usage

from tsseg.metrics import F1Score, StateMatchingScore

# Change-point evaluation
f1 = F1Score(margin=5)
result = f1.compute(y_true_cps, y_pred_cps)
print(result["f1"])

# State-labelling evaluation
sms = StateMatchingScore()
result = sms.compute(y_true_labels, y_pred_labels)
print(result["score"])

See Getting started for end-to-end examples that combine detectors and metrics.