Open ServetKocak opened 2 years ago
Hi, have you figured out this? I am also quite confused about this. Thanks a lot!
Hi @ly1112, I contacted to author of the paper and he said "In the paper we normalized the anomaly scores between 0 and 1.0 and then tested 10 000 thresholds (0.0001 , 0.0002 , ....)."
I also found another USAD application and they did the following:
Good luck!
Hi @ly1112, I contacted to author of the paper and he said "In the paper we normalized the anomaly scores between 0 and 1.0 and then tested 10 000 thresholds (0.0001 , 0.0002 , ....)."
I also found another USAD application and they did the following:
Good luck!
Hi. thank you so much for your helpful reply! So they are doing a grid search on all possible threshold, which will be quite time-comsuming..... As for your another find, they are more easy to impliment. May I ask, can you share the link of these application? And did you reproduce the results of the paper based on the new application on threshod? Thanks a lot!
Hi @ly1112, I contacted to author of the paper and he said "In the paper we normalized the anomaly scores between 0 and 1.0 and then tested 10 000 thresholds (0.0001 , 0.0002 , ....)."
I also found another USAD application and they did the following:
Good luck!
hi @ly1112 , where did you find this implementation? Thanks for your help
Hi @ly1112 @ServetKocak @dadodimauro
Can you share the complete code with the best implementation? The code published by the original author is not complete.
Thanks for your help!
Hi @ly1112 @ServetKocak @dadodimauro
Can you share the complete code with the best implementation? The code published by the original author is not complete.
Thanks for your help!
Hi, I can only suggest you to look at my public repo called anomaly detector where I implemented the USAD algorithm to find anomalies in an oracle database.
I did my masters thesis work on that, unfortunately I don't remember well the specifics, here you can find my thesis, I hope you find it useful thesis
Cheers
Hi @manigalati , I am wondering how did you implement the threshold as AUC score? Isn't the AUC score is the kinda accuracy of the model. If you check the maximum value of y_pred, it is lower than the AUC score thus In the paper, they don't explain how did they set the threshold.