search

liyaguang / DCRNN

Implementation of Diffusion Convolutional Recurrent Neural Network in Tensorflow
MIT License
1.19k stars 394 forks source link

Abnormal Train Loss #38

Closed simmonssong closed 5 years ago

simmonssong commented 5 years ago

Hi, thanks for your code. I got an abnormal training loss under default config. Although validating error is convergent, training loss goes down at first but it keeps growing after minimal.

liyaguang commented 5 years ago

This behavior is expected due to the scheduled sampling. In this process, the training task keeps getting hard, and finally converges to the testing/validation task. You may refer to Section 2.3 in the paper for more information.

simmonssong commented 5 years ago

This behavior is expected due to the scheduled sampling. In this process, the training task keeps getting hard, and finally converges to the testing/validation task. You may refer to Section 2.3 in the paper for more information.

I got it. Thanks a lot.