search

uci-cbcl / UFold

MIT License
58 stars 26 forks source link

model.eval() mode #18

Open giorgiobini opened 1 year ago

giorgiobini commented 1 year ago

Hello,

I see you have BatchNorm2d layers, so the output of the model with change accordingly to the train() and eval() mode you use. I was wondering why you haven't tested the model with the eval() mode.

Thank you so much in advance!

sperfu commented 1 year ago

Hi,

We trained/evluated our model with batch size 1. Different sequences have very different running statistics, so we don't want the model to use the previously stored statistics.