zhiguowang / BiMPM

BiMPM: Bilateral Multi-Perspective Matching for Natural Language Sentences
Apache License 2.0
438 stars 150 forks source link

The results of test set and verification set differ greatly. What is the problem? #62

Open 13593574504 opened 3 years ago

13593574504 commented 3 years ago

Hello, I used your Quora data set and the corresponding configuration to train 10 EPOCHs on the model. During the whole training process, the ACC of the verification set performed well, about 87%. However, when I tested the test set with the trained model, the ACC decreased significantly, fluctuating between 52% and 58%, which made me very perplexed. May I ask what the reason is?