search

thuiar / Self-MM

Codes for paper "Learning Modality-Specific Representations with Self-Supervised Multi-Task Learning for Multimodal Sentiment Analysis"
MIT License
176 stars 35 forks source link

About the result on MOSI #4

Closed lzjjeff closed 3 years ago

lzjjeff commented 3 years ago

你好~我按照README上的步骤试着在自己的电脑上跑了一下代码，在mosi上的结果和论文里的有些差距：

Has0_acc_2=82.97 Has0_F1_score=82.99 Non0_acc_2=84.6 Non0_F1_score=84.57 MAE=72.52 Corr=79.44

因为论文中提到的结果是5次结果的平均，这样的差距让人有点意外。

除了transformers==4.4.2以外，requirements.txt内所要求的环境都满足，另外我的cuda版本为10.2。

请问我需要更改模型参数或者训练参数吗？能否提供更为详细的环境信息呢？

期待回复~

iyuge2 commented 3 years ago

Please see Issue 2. Thanks~

lzjjeff commented 3 years ago

Please see Issue 2. Thanks~

好的，感谢~

leijue222 commented 2 years ago

@lzjjeff Have you achieved results comparable to those of the paper? My result is similar to yours, but this result is still a bit worse than the paper result.

lzjjeff commented 2 years ago

@lzjjeff Have you achieved results comparable to those of the paper? My result is similar to yours, but this result is still a bit worse than the paper result.

Not yet. I re-run it with the new config file updated in May but got similar results to my previous one.

leijue222 commented 2 years ago

Unfortunately, we lost the original parameters in our paper when we re-run all models and datasets. But you can try the following parameters, which can get comparable or better results than our work in AAAI 2021.

Too sad... I tried a lot but couldn't get results comparable to the paper...