search

hiyouga / Dual-Contrastive-Learning

Code for our paper "Dual Contrastive Learning: Text Classification via Label-Aware Data Augmentation"
https://arxiv.org/abs/2201.08702
MIT License
154 stars 27 forks source link

Some questions with baselines #7

Open TaoCesc opened 2 years ago

TaoCesc commented 2 years ago

Your work is very good and effective. But I have some questions about the baseline approach. I tried different hyperparameters to adjust supervised contrastivelearning or unsupervised contrastive learning to fine-tune BERT, and then to classify. But I've never been able to do anything better than just Cross-Entropy. I wonder what I didn't take into account? I've seen a lot of papers that contrastive learning can help improve classification results, but here I always get the opposite. Maybe I want to know the hyperparameters you set when you ran the comparison.

wangqian97 commented 2 years ago

Hello, I tried to use author code, but ACC is stable to 50%, I want to know how you do the training.

TaoCesc commented 2 years ago

Hello, I tried to use author code, but ACC is stable to 50%, I want to know how you do the training.

Hi，When I reproduced the experiment, I did not modify other parameters. Basically, I used the parameters given by the author, and only modified the size of batch_size. I can reproduce the author's results and get good results on my own dataset, you may need to recheck the parameters or download a new code to try it out.

DominicSlw commented 2 years ago

Your work is very good and effective. But I have some questions about the baseline approach. I tried different hyperparameters to adjust supervised contrastivelearning or unsupervised contrastive learning to fine-tune BERT, and then to classify. But I've never been able to do anything better than just Cross-Entropy. I wonder what I didn't take into account? I've seen a lot of papers that contrastive learning can help improve classification results, but here I always get the opposite. Maybe I want to know the hyperparameters you set when you ran the comparison.

Hi, can you explicitly explain what your hyperparameters are when using supervised contrastive learning to fine tune the baseline? I get the same result with you. Thank you!