Your system seems to have great performances based on what you reported in the paper. Now I would like to apply your system to my custom dataset.

Do you make any additional changes to simpletransformers library except those shown below? It is possible to only include these custom code and make it work with simpletransformers installed through pip install simpletransformers?

I am very interested in your model and am currently trying to replicate your results for the ag_news dataset. I finally managed to run train_ag.py which saves the pytorch model and the configs to the output_dir. Unfortunately, I can't figure out what the next steps are to get the anomaly scores. Can it be that a file is missing in which the evaluation is implemented?