What are the "Leaks features" and how is the accuracy measured?

[iuria21]

Hi, first thanks for your repo! I have two questions:

1- I understand that "leaks features" are like common characters between sentences in each pair or something like this? I I'm trying to train the model with vector similarity directly instead of sentence similarity, do you guess I can remove the leaks part from the model?

2- How is the accuracy measured? I guess that the sigmoid is outputting a number between 0 and 1, so how is the accuracy measured, considering 1 if the value is >0.5 and 0 else?

Thanks!

[Ahmed-M0hamed]

I think of course you can it is depend and if your dataset set has other features you can concatenate them instead of the leaks features