thuiar / Cross-Modal-BERT

CM-BERT: Cross-Modal BERT for Text-Audio Sentiment Analysis(MM2020)
99 stars 26 forks source link

Fintune problem #3

Closed cyZhu98 closed 2 years ago

cyZhu98 commented 2 years ago

非常感谢作者开源。我注意到你们的代码没有解冻最后一层分类层,我加上对比了下,精度稍微低了一点,不过这有可能是因为超参不合适。你们这样做是因为解冻分类层效果会更差,还是说有其他什么讲究呢? i wonder why you didn't unfreeze the classifier layer in run_classifier.py line129 ? I got a slightly lower accuracy when i unfreezed 'classifier' which may due to the unmatched hyperparameters, or is there something else

Kaicheng-Yang0828 commented 2 years ago

您好,感谢您的关注。事实上并没有什么讲究,您可以不用过于纠结,他们效果应该是差不多的,您可以多尝试下其他超参数。

cyZhu98 commented 2 years ago

好的,谢谢!