How to monitor tensorboard logged scalar in modelcheckpoint?

[ghost]

If I save scalar log like this,

self.logger.experiment.add_scalars(‘loss/nll’, {‘train’: trainloss, ‘valid’: validloss})

How to monitor valid loss in Modelcheckpoint?

[justusschock]

you need to call it with self.log instead (we automate the rest for you), so that we are aware of you logging it :)

[ghost]

If one wants to group them and plot them into single figure to compare, one should log like i mentioned as #3932 (comment)

So there is no way to monitor if one logs with self.logger.experiment?

[justusschock]

No, because we actually cache the values for monitoring and if you directly use self.logger.experiment you bypass our mechanism there which results in us not knowing about that value at all