What's the meaning of '2876934.2 / 1281167 * self.data_len' when calculating instance_Z?
walkingwindy opened this issue · 2 comments
walkingwindy commented
Thanks for this great work.
However, there still exits a problem making me confused, that is how to calculate the Z in NCE loss.
In the implementation of the code, I found that Z is defined as '2876934.2 / 1281167 * self.data_len',
I wondering what's the meaning of 2876934.2 and 1281167
THANKS FOR ANY HELPFUL ADVICES:)
chengxuz commented
Hi, please check explanations here: neuroailab/LocalAggregation-Pytorch#1, sorry for the confusion.
walkingwindy commented
Oh!I got it. Thanks for your reply!