README.md

LSTM-AD

Citekey	MalhotraEtAl2015Long
Source code	own
Learning type	semi-supervised
Input dimensionality	multivariate

Dependencies

• python 3
• pytorch

Notes

LSTM-AD outputs anomaly scores for windows. The results require post-processing. The scores for each point can be assigned by aggregating the anomaly scores for each window the point is included in.

You can use the following code snippet for the post-processing step in TimeEval (default parameters directly filled in from the source code):

from timeeval.utils.window import ReverseWindowing
# post-processing for LSTM-AD
def post_lstm_ad(scores: np.ndarray, args: dict) -> np.ndarray:
    window_size = args.get("hyper_params", {}).get("window_size", 30)
    prediction_window_size = args.get("hyper_params", {}).get("prediction_window_size", 1)
    return ReverseWindowing(window_size=window_size + prediction_window_size).fit_transform(scores)