Compare hyperparameters in Tensorboard

I want to be able to group runs by hyperparameter. For example, I should be able to select “lr” in tensorboard and see the plot for the loss where each line is the mean of all runs with the same learning rate (lr).

This seems to be a very common way of analyzing training plots for hyperparameters, but it’s missing.

*** You could potentially do this with regex and putting the hyperparameter names as the run directory name… but a folder name is limited at 255 characters!!! It’s not a good solution.

Bumping this. I think people may have hyperparameter tuning pipelines that must work better than putting the parameters in the folder titles and regexing in tensorboard?