etna.analysis.get_anomalies_density#

get_anomalies_density(ts: TSDataset, in_column: str = 'target', window_size: int = 15, distance_coef: float = 3, n_neighbors: int = 3, distance_func: Literal['absolute_difference'] | Callable[[float, float], float] = 'absolute_difference', index_only: bool = True) → Dict[str, List[Timestamp] | List[int] | Series][source]#

Compute outliers according to density rule.

For each element in the series build all the windows of size window_size containing this point. If any of the windows contains at least n_neighbors that are closer than distance_coef * std(series) to target point according to distance_func target point is not an outlier.

Parameters:

  • ts (TSDataset) – TSDataset with timeseries data

  • in_column (str) – name of the column in which the anomaly is searching

  • window_size (int) – size of windows to build

  • distance_coef (float) – factor for standard deviation that forms distance threshold to determine points are close to each other

  • n_neighbors (int) – min number of close neighbors of point not to be outlier

  • distance_func (Literal['absolute_difference'] | ~typing.Callable[[float, float], float]) – distance function. If a string is specified, a corresponding vectorized implementation will be used. Custom callable will be used as a scalar function, which will result in worse performance.

  • index_only (bool) – whether to return only outliers indices. If False will return outliers series

Returns:

dict of outliers in format {segment: [outliers_timestamps]}

Return type:

Dict[str, List[Timestamp] | List[int] | Series]

Notes

It is a variation of distance-based (index) outlier detection method adopted for timeseries.