How is metric used in optimization?

[dr-upsilon]

Hi everyone,

I have been digging around the code for a while and I cannot get my head straight on how metrics are used in pytorch forecasting.

I saw loss is called via log() in step() for each batch. Could somebody explain this to me? So in PF is the metric calculated at batch level and possibly reset? Does it make sense to take average at epoch level?

Thank you.