search

XueFuzhao / GDPNet

38 stars 13 forks source link

any plan to release the code of SimpleRE？ #2

Closed nttmac closed 3 years ago

nttmac commented 3 years ago

Hello, author! I read another related paper, "An Embarrassingly Simple Model for Dialogue Relation Extraction". Do you have any plan to release the code?."

XueFuzhao commented 3 years ago

Yes, absolutely. We are cleaning the code and try to extend the experiments. I'm also working on a few other deadlines. So there is no clear timeline about that. Actually, the SimpleRE is very very simple and easy to implement. I believe it would not spend too much time to reimplement it on your side. Certainly, if you have no urgent deadline to use the SimpleRE model, you can wait for our code. I promise we will release it ASAP.

nttmac commented 3 years ago

Yes, absolutely. We are cleaning the code and try to extend the experiments. I'm also working on a few other deadlines. So there is no clear timeline about that. Actually, the SimpleRE is very very simple and easy to implement. I believe it would not spend too much time to reimplement it on your side. Certainly, if you have no urgent deadline to use the SimpleRE model, you can wait for our code. I promise we will release it ASAP.

Actually, I have implemented the SimpleRE without RRG. The average and standard deviation of F1 scores are 63.32 and 2.10 respectively, with a gap between the results in your paper 65.5 (0.7).

XueFuzhao commented 3 years ago

Thank you for your interest! Please follow the hyperparameters stated in the paper, and add some dropout before classification. Since DialogRE is not very large, the std would be high if we do not add dropout. BTW, training more epochs may be helpful to get a more stable result.

nttmac commented 3 years ago

Thank you for your interest! Please follow the hyperparameters stated in the paper, and add some dropout before classification. Since DialogRE is not very large, the std would be high if we do not add dropout. BTW, training more epochs may be helpful to get a more stable result.

Yeah! This time, I set the batch size 6, and the results are consistent with those in your paper! thank u for your help and your patience!

XueFuzhao commented 3 years ago

So happy to hear that!!! And Yes, we have to set the batch size as 6 in SimpleRE because we have so many labels in each dialogue. Welcome further discussion and collaboration if you are interested in this topic in the future!