Request for information regarding TS classification dataset labelling #2
Description
Hello @boniolp @sylligardos
In the coming days I would like to try your model selection algorithms, but on my own TS datasets.
Could you please advise how to create the TS classification dataset needed for the Model Selection methods in MSAD ?
From my understanding, this repository is also relying on another one (https://github.com/TheDatumOrg/TSB-UAD),
which is about the training and evaluation of individual univariate TSAD algorithms on a given collection TS datasets. I assume that this is needed, since it will provide the anomaly prediction labels for a given pool of detectors of interest.
However, I lack clarity on the steps and scripts needed in order to reach my goal.
Questions:
- Are there some documentation/examples of how to integrate a new dataset to TSB-UAD ?
- When evaluating an individual detector in TSB-UAD, are the detection metrics saved in the format needed by MSAD?
- Also is there a minimum number of detectors from which I have to gathers metrics logs ?
- Any particular post-processing steps needed for MSAD?
I would really appreciate your guidance.
Thank you so much!
Best regards
René