Congratulations and Questions

[BraudoCC]

Congratulations on achieving such great results! I encountered a question during my study:

I found that the test set of the experiment contains data from the training set. Is this kind of setup widely applied? Because I noticed that if I change the test set to the last 30% of the data, there are no anomaly samples in some time-series data.

Also, have you considered open-sourcing the original data?

[qhliu26]

Hi, thank you for your interest in our work! Please refer to the following discussion for details on the training/test data splitting:
#6 (comment).

The original dataset can be accessed here: https://www.thedatum.org/datasets/TSB-UAD-Public-v2.zip
(It contains 29 datasets and 3,427 time series.)

Let us know if you have any further questions!