search

W-lw / PCLC

Prototypical contrastive learning and label confusion for cross-domain slot filling.
10 stars 0 forks source link

Can not reproduce the results ? #3

Open LittleSummer114 opened 2 years ago

LittleSummer114 commented 2 years ago

I have read your answers "Thank you for your reminder. In this version of the code in this repository, we have simplified some of the hyperparameters to uniform hyperparameters, so the results for some domains may differ. We'll recalibrate the code later." When will you release the code ?

LittleSummer114 commented 2 years ago

Hi, I found if I use the models you release, the results can be reproduced. However, I found that the performance on seen and unseen slots are higher than you reported! Is there anything needed to be modified ? e.g. Evaluate on RateBook domain unseen slots. Final slot F1 score: 29.3069. (18.81)