gluonts.nursery.anomaly_detection.filters module#

gluonts.nursery.anomaly_detection.filters.fill_forward(a: numpy.ndarray, fill_start=None) → numpy.ndarray[source]#

Forward fill an array.

If fill_start is not None, then the NaNs in the beginning of the array will be filled with fill_start.

gluonts.nursery.anomaly_detection.filters.labels_filter(a: numpy.ndarray, num_open: int, num_clear: int, initial_label: int = 0, forward_fill: bool = False)[source]#
gluonts.nursery.anomaly_detection.filters.n_k_filter(a: numpy.ndarray, num_open: int, num_clear: int, num_open_suff: Optional[int] = None, num_clear_suff: Optional[int] = None, initial_label: int = 0, forward_fill: bool = False) → numpy.ndarray[source]#

Implements the (N, K)-filter that opens/closes anomalies based on observing K or greater labels in the last N time steps.

Parameters

  • a – input array, type boolean

  • num_open – number of time steps to look back for opening an anomaly

  • num_clear – number of time steps to look back for clearing an anomaly

  • num_open_suff – number of positive labels in the lookback period sufficient to open an anomaly. defaults to num_open.

  • num_clear_suff – number of positive labels in the lookback period sufficient to clear an anomaly. defaults to num_clear.

  • initial_label – 1 if the initial state is an anomaly, 0 if not. default 0.