Anomaly detection with an array of labels?

[petteriTeikari]

Hi,

I am new to aeon and stumbled on this as UniTS suggested converting my own data to .ts format. I could not quickly see how can you supply an array of labels to anomaly detection?

e.g. in the tutorial:
https://github.com/aeon-toolkit/aeon-tutorials/blob/main/ECML-2024/Notebooks/part6_anomaly_detection.ipynb

or in the desired tutorial issue:
#1960

I have for example (16, 1, 1981) input data array with the 16 subjects being totally independent, and having their outliers happening at different times and cannot give a y size of (1981,) but would like to give (16,1981) type of shape.

Or do I simply just concatenate the array to (1,1,1981*16) shape with y with shape (1981*16)

Thanks

[SebastianSchmidl]

Is this question about (1) storing anomaly detection data, including the anomaly scores/labels, to disk using the .ts-file format, or (2) using estimators from aeon.anomaly_detection to detect anomalies in multiple time series?

1. You cannot store the anomaly labels/scores (of shape (n_timesteps,)) in a .ts-file, we currently use the TimeEval file format.
2. The estimators in aeon.anomaly_detection are all series estimators and do not support collections. You need to call the estimator's predict for each time series individually.