search

songyouwei / ABSA-PyTorch

Aspect Based Sentiment Analysis, PyTorch Implementations. 基于方面的情感分析，使用PyTorch实现。
MIT License
2.03k stars 528 forks source link

Request for K-fold cross-validation support #32

Closed ErfanThinker closed 5 years ago

ErfanThinker commented 5 years ago

Is it possible to add K-fold cross-validation support to the code?

songyouwei commented 5 years ago

Most ABSA papers do not use K-fold cross-validation in their research, and there is no validation set. I think it would be a good idea to add K-fold cross-validation support in a new branch.

ErfanThinker commented 5 years ago

Actually it can be used on bing liu's datasets as used in https://link.springer.com/article/10.1007/s10586-017-1096-9 (10-fold cross-validation)

jiangqn commented 5 years ago

Are the hyper-parameters finetuned on the test set?

songyouwei commented 5 years ago

@ThomasK427 Codes for hyper-parameter tuning are not included in this repo. You can randomly split 20% of training set as the DEVset, and tune hyper-parameters on this DEVset.

songyouwei commented 5 years ago

Supported in this commit.