detection

Pass the absolute model score through as the anomaly score.

Decide with a threshold derived from the observed score distribution.

One decided anomaly, possibly spanning an interval.

Detector decision output: events plus the labeled score table.

Compose model, scorer, optional calibrator, and detector into one method.

TimeSeries scorer contract, distinct from legacy scored-table components.

Continuous anomaly scores with explicit alignment information.

Convert continuous scores into decisions or anomaly events.

Convert model outputs into standardized anomaly scores.

Score anomalies from distances to a learned normality boundary.

Cauchy negative log likelihood using robust median and half-IQR estimates.

Decide by detecting distribution shifts in the score sequence.

Combine the scores of several child scorers into one score column.

Convert continuous scores into decisions or anomaly events.

Signed per-component difference ``actual - predicted`` (no fitting).

Score anomalies directly on the observations without a model output.

Aggregate raw decisions into interval events with severity.

Exponential negative log likelihood for nonnegative prediction samples.

Gamma negative log likelihood with method-of-moments sample parameters.

Gaussian negative log likelihood using sample mean and standard deviation.

Wrap a detector callable or estimator and normalize binary labels.

Wrap a scoring callable or estimator with explicit anomaly direction.

Learn Tukey fences ``Q1 - scale*IQR`` and ``Q3 + scale*IQR``.

Euclidean distance to the closest fitted normal-window cluster center.

Laplace negative log likelihood using sample median and absolute deviation.

Score anomalies from estimated likelihoods.

Negative log likelihood fitted separately to prediction samples per point.

Absolute component residuals or their vector norm at each timestamp.

Raised when component capabilities do not match the observation set.

Poisson negative log likelihood for nonnegative integer observations.

Adapt a PyOD-compatible estimator's fit/decision_function to windows.

Learn lower/upper per-component quantiles from explicit training scores.

Flag the top quantile of a score column as anomalous.

Score anomalies from reconstruction errors.

Score anomalies from latent-space distances or densities.

Score anomalies from prediction or forecast residuals.

Alignment level of an anomaly score.

Convert model outputs into standardized anomaly scores.

Decide with a fixed, configured threshold.

Decide with statistical tests or control-chart rules.

Flag anomalies whenever a score column crosses a fixed threshold.

Flag scores strictly below or above explicit component-wise bounds.

One-dimensional Wasserstein distance from each window to training values.

Fit and score sliding windows without crossing series boundaries.

Normalize a model score column with a z-score.