Results of another theia dataset

[Garvit2543]

I have tried this model with no changes in code and tried implementing from scratch for theia fataset tr1-trace-e3-official-1r.json and same ground truth file of theia from tracethrea
But I'm not getting results as expected at evaluation

TN and FN as 0
Can u please suggest me whats the mistake I'm doing.
Do i need to update code or some file

[SharedyX]

Hello bro, I have the same question. Have you found a solution?

[Jimmyokok]

1. MAGIC is an unsupervised deep-learning based model, which requires as much (benign) training data as possible. Trying increasing more json files (e.g. all theia-1r json files) for training and see if AUC increases.
2. Remember to unset the pre-defined detection threshold. If the threshold is too harsh, zero TN and FN can be observed.