How to log the learning rate with pytorch lightning when using a scheduler?

W&B Help
1

I’ve been trying to find some documentation, I don’t want to save all the hyperparameters each epoch, just the learning rate.
Would be so great if you can help me out.

Cheers,

Oli

2

Hi Oli,

Just double checking, are you talking about running sweeps?

Cheers,
Artsiom

3

Hi there, I wanted to follow up on this request. Please let us know if we can be of further assistance or if your issue has been resolved.

4

Hi Oliver, since we have not heard back from you we are going to close this request. If you would like to re-open the conversation, please let us know!

5

I’m also wondering how this is done! Whether within a sweep configuration or not - when using a lr scheduler, I am trying to track the lr at epoch during training, as it is now dynamic. Even within a sweep, you will have some initial lr determined during the sweep, but it will not stay constant for the duration of training.

edit:

The example on the lightning site here worked for me:

>>> from pytorch_lightning.callbacks import LearningRateMonitor
>>> lr_monitor = LearningRateMonitor(logging_interval='step')
>>> trainer = Trainer(callbacks=[lr_monitor])

Passing the WandBLogger to the trainer I see my lr is logged on the wandb dashboard.

2 Likes
6

Hi there, yes,
sorry for the late reply, I didn’t get any notification from your answers, but now I got one from Chris’s.

What worked for chris also worked for me.

2 Likes
7

This topic was automatically closed 60 days after the last reply. New replies are no longer allowed.