left_stampi

Left STAMPi

Citekey	YehEtAl2016Matrix
Source Code	stumpy
Learning type	unsupervised
Input dimensionality	univariate

This approach uses the left matrix profile (LMP). That is extended with every new point in the stream. The already existing LMP entries are not updated.

Output Format

The output will be an anomaly score for every input data point

Dependencies

• python 3
• numpy
• pandas
• stumpy

Notes

from timeeval.utils.window import ReverseWindowing
# post-processing for left_stampi
def post_left_stampi(scores: np.ndarray, args: dict) -> np.ndarray:
    window_size = args.get("hyper_params", {}).get("anomaly_window_size", 50)
    n_init_train = args.get("hyper_params", {}).get("n_init_train", 50)
    if window_size > n_init_train:
        print(f"WARN: anomaly_window_size is larger than n_init_train. Dynamically fixing it by setting anomaly_window_size to n_init_train={n_init_train}")
        window_size = n_init_train
    if window_size < 3:
        print("WARN: anomaly_window_size must be at least 3. Dynamically fixing it by setting anomaly_window_size to 3")
        window_size = 3
    return ReverseWindowing(window_size=window_size).fit_transform(scores)