Logic behind presence check in text_explain

[hshany]

Thanks for the wonderful tool. I'm testing the library by following the tutorial with some of my own prediction models. I noticed that the text explanation is a filtered list of the descriptors displayed on the images. Based on the code, part of the filter is a check on the presence_threshold. What is the logic behind this check? From the code one can infer that in the case where present=nbases=1, the condition "if present / nbases < (1 - presence_thresh) and v < 0" will always fail for negatively contributing descriptors with finite presence_thresh, which means negatively contributing descriptors will always be excluded from the text description. Is there a rational supporting such exclusion?

https://github.com/ur-whitelab/exmol/blob/67873897bbcd2deb60c69a5116bc7feea69fdefa/exmol/exmol.py#L1462C1-L1475C21