Running Average of my accuracy, losses etc.

[karanchahal]

What is your question?

I want my tqdm logger to show me a history of my training on the terminal. Right now, when a epoch ends, all data for it is scrubbed from the command line and the new epoch data is shown.

Also I want to see the running accuracy of my network and a running average of my loss on the tqdm bar. How should I go on about doing that ?

What have you tried?

I have looked at the docs and logging but am unable to figure out how to modify the tqdm logger, more so maintain a running average

What's your environment?

• conda version: latest
• PyTorch version : 1.3
• Lightning version : pip install pytorch-lightning at the date of this issue
• Test-tube version: came boot strapped with lightning.

I installed everything on the date of this issue.

[s-rog]

Use tensorboard!

For running averages you have to implement the logic in training step

[karanchahal]

Thank you, yes I've figured it out now !