'main/accuracy and validation/main/loss' value is nothing and not printed during training.

[Kevin Ryu]

Hi,

A validation data was prepared to validate training and validation, and an Evaluator was added to train code. The code does not seem to be a problem compared to the example.

The validation loss calculation code should be directly implemented and added to the model code? If I have to distinguish the main loss and validation loss calculation code in the model code, is there a related example? I wonder if there is only one loss calculation block in the model code, not the main and validation, and whether the validation loss is calculated automatically when the Evaluator operates. But 'main/accuracy and validation/main/loss' value is nothing and not printed during training. And no value is displayed in 'main / administrator' when executing 'PrintReport' regardless of adding validation. When adding and executing validation related code, an error occurs that the 'validation / main / loss, validation / main / verification' value is not displayed and the 'dtype' does not appear normal after the second epoch and the learning ends. Please understand the difficulty of sharing the entire code, and please give feedback to those who have similar experiences. 

Of course, there is no loss curve plot image at all even though the function is coded.  What causes it?

Training environment:

Ubuntu 18.04 LTS

chainer 7.1.0

Nvidia Cuda 10.1

three titan xp graphic card.

That's all.