daxs.filters: Filters for anomaly detection#
The module provides functions for filtering data.
- daxs.filters.hampel(data: npt.NDArray[np.float64], window_size: int | None = None, threshold: float = 3.5, axis: int = 0, k: float = 1.4826)[source]#
Outliers detection using the Hampel filter.
More details about the filter can be found here:
- Parameters:
data – Input data.
window_size – Size of the sliding window.
threshold – Threshold for outlier detection expressed in number of standard deviations. Iglewicz and Hoaglin [1] suggest using a value of 3.5, but larger values are often needed to avoid removing data points from a noisy signal.
axis – Axis along which the detection is performed.
k – Scale factor for the median absolute deviation. The default value is 1.4826, which is the scale factor for normally distributed data.
- Returns:
Mask identifying the outliers, and the rolling window median.
- Return type:
np.NDArray[np.bool_]
References