search

yang-yk / NP-RepMet

27 stars 8 forks source link

loss in the code is not the same as in the paper #4

Open kyrawu327 opened 3 years ago

kyrawu327 commented 3 years ago

Hi, thank for the nice work and the idea of restoring negative information is quite inspiring. However, I find that the loss in the code is not the same as in the paper, am I missing something? or the code is further optimized?

Looking forward to your kind reply.

yang-yk commented 3 years ago

Hi, thank for the nice work and the idea of restoring negative information is quite inspiring. However, I find that the loss in the code is not the same as in the paper, am I missing something? or the code is further optimized?

Looking forward to your kind reply.

Yes. The loss is further optimized with the cosine distance. Thanks！