should we use an averaged model for marginal footprint analysis?

[Brawni]

Hello again!

For marginal footprinting should i use no bias models for each fold and then average the h5 footprints file or there is a way to average nobias model h5 files from the different folds beforehand? And related to this question, I'm currently training nobias models using bias models trained on corresponding folds. Is this the way to do?

Thanks so much!

[mfansler]

"...I'm currently training nobias models using bias models trained on corresponding folds. Is this the way to do?"

I came here with the same question. Specifically, the section on bias model training refers to "a good and complete bias model". This hints that one could generate a single bias model that could, e.g., be used across folds of bias-factorized trainings. I think the discussion of expected use of bias models could be benefit from some elaboration.