ruiking04 / COCA

Deep Contrastive One-Class Time Series Anomaly Detection
30 stars 9 forks source link

Multiple Features #30

Open marciahon29 opened 1 month ago

marciahon29 commented 1 month ago

Hello,

Currently, COCA accepts datasets that are univariate. Is it possible to expand to multivariate?

Thanks, Marcia

ruiking04 commented 1 month ago

Sorry, the original COCA does not perform well on multivariate time series. We guess that this may be because the model is not complex enough. You can try deepening the network layers and replacing LSTM with the Transformer.

marciahon29 commented 1 month ago

Thanks for the tip!

On Tue, Sep 10, 2024 at 5:33 AM RuiKing @.***> wrote:

Sorry, the original COCA does not perform well on multivariate time series. We guess that this may be because the model is not complex enough. You can try deepening the network layers and replacing LSTM with the Transformer.

— Reply to this email directly, view it on GitHub https://github.com/ruiking04/COCA/issues/30#issuecomment-2340154689, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFOIJPXNWO4H6FOTFEPWXDTZV24FPAVCNFSM6AAAAABNLLPYBGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDGNBQGE2TINRYHE . You are receiving this because you authored the thread.Message ID: @.***>