How to log hyperparameters in Tensorboard HPARAMS tab?

[00sapo]

Hello,
from the documentation, it's not completely clear to me how to log hyperparameters in Tensorboard.

The documentation is at https://lightning.ai/docs/pytorch/stable/extensions/logging.html#logging-hyperparameters

It's not clear if and how torch.utils.tensorboard.writer.SummaryWriter.add_hparams() should be called. Moreover, I'm using self.logger.log_hyperparams(self.hparams, {"hp/metric_1": 0, "hp/metric_2": 0}) as in the example, but all I get is this

I would expect Start Value to be 0, at least. The HPARAMS tab instead shows the usual empty hyperparameters warning.

Can somebody show me a simple example for using the HPARAMS tab?

[adosar]

@00sapo I have opened a similar discussion here. Do you get "No hparams data was found."?

[00sapo]

Yep

[leSullivan]

Any solutions ?

[mazurkin]

#19960 (comment)

[MukundaKatta]

HPARAMS needs two calls. log_hyperparams(self.hparams, {"hp/val_acc": 0.0}) registers the hyperparams and placeholder metrics, then self.log("hp/val_acc", acc) inside validation_step with the exact same hp/... name produces the scalar history TB needs. Two gotchas: names must match prefix-for-prefix (hp/val_acc in log_hyperparams is not the same as val_acc in self.log), and call log_hyperparams once (in on_train_start) not every step, otherwise you get duplicate HPARAMS entries.