The paper claims that the threshold was 0.5 however in the code, specifically in the predict method in document_bert.py, the threshold's default value is 0. In the fit method, the predict function is called with no threshold argument so the threshold is 0. I just thought this could potentially be an error in the code.
The paper claims that the threshold was 0.5 however in the code, specifically in the predict method in document_bert.py, the threshold's default value is 0. In the fit method, the predict function is called with no threshold argument so the threshold is 0. I just thought this could potentially be an error in the code.