Better performance when reproduce paper

shun-zheng / Doc2EDAG

MIT License

342 stars 72 forks source link

Better performance when reproduce paper #26

Open YuanEric88 opened 3 years ago

YuanEric88 commented 3 years ago

Hi, when I reproduced the code with the same hyparameter setting, I got the result better result with an non-negligible scope compared to that shown in the paper: mine is around 77.8%(the result in paper is only 76.2) Is it reasonable?

shun-zheng commented 2 years ago

It is reasonable since the results reported in the paper was based on the two-years-ago PyTorch and CUDA backends.