predprob = float(expit(torch.mean(logit(subseq))))

[igortru]

Hello,Markus!
I see minor issue with logit, if subseq contain 0.0, logit return -inf , if subseq contain 1.0 logit return inf,
as result mean will be corresponding inf
and expit automatically return 0.0 or 1.0
is it intentional? one 0.0 or 1.0 make "mean" useless,

torch.logit has additional attribute eps which allow exclude such situations.

https://pytorch.org/docs/stable/special.html#torch.special.logit

what do you think?