search

henryhungle / NADST

Code for the paper Non-Autoregressive Dialog State Tracking (ICLR20)
MIT License
44 stars 5 forks source link

Is it possible to share your trained model? #6

Closed YourThomasLee closed 4 years ago

YourThomasLee commented 4 years ago

Hi! I am a master candidate doing the research of DST. I have downloaded your code and trained new model. So many times the result of test is 

"Joint Acc 0.4767915309446254 Slot Acc 0.9717969598263123 F1 0.8849829769540185
Oracle Joint Acc 0.6999185667752443 Slot Acc 0.9448231540611014 F1 0.9881572824569892"

As you know, the joint accuracy result reported in paper was 0.49. 1. I want to know is there any trick in training? if not, could you possibly share your trained model with us? 2. if possible, can you share your experience of generating best model. It will help me a lot in my research. Thank you!

YourThomasLee commented 4 years ago

After many times of running program, I have got the best model as reported in paper. Thanks a lot!