Tabular data regression,direction of validation loss change does not match with the direction of RMSE change

I am using tabular model for regression problem. I use default flatten MSE loss as the loss function and RMSE as the metric. One thing that I have noticed is that sometimes the direction of the validation loss change and that of RMSE change does not match with each other. As the image showed as follows, from the epoch 2 to epoch 3, the valid_loss is decreased from 78.946 to 77.521, while the RMSE is increased from 7.181 to 7.237.

Does anyone have idea about how the phenomenon occurs? Thanks a lot!

1 Like

I am seeing the same results, did you figure out why this is the case?