Logging multiple metrics to hparams tab in tensorboard

[adobles96]

Hi,

I've been trying to log multiple metrics to the hparam tab in tensorboard but haven't been able to do it. This is what I'm doing (mostly following the documentation):

class MyModule(pl.LightningModule):
    def __init__(self, config: Dict):
        self.save_hyperparameters(config)
        self.metrics = [
            'hp/best_val_loss',
            'hp/best_val_acc',
        ]
        ...
    
    def on_train_start(self):
        self.logger.log_hyperparams(self.hparams, {k: 0 for k in self.metrics})

    def validation_epoch_end(self, val_outputs):
        metrics = calc_aggregated_metrics(val_outputs)
        for metric, value in metrics.items():
            self.log(metric, value)  # here metrics corresponds to the metrics in self.metrics
            

tb_logger = pl_loggers.TensorBoardLogger('logs/', default_hp_metric=False)
M = MyModule(config)
trainer = pl.Trainer(logger=tb_logger)
trainer.fit(M, train_loader, val_loader)

When looking at tensorboard, the 'hp/*' metrics show up fine in the scalars tab, but in nothing shows up in the hparams tab. What's interesting is that if I comment out the self.log line in validation_epoch_end then I do see hparams in the hparams tab (but obviously not the metrics), so there seems to be a conflict between logging the metrics and seeing the hyperparameters in the hparams tab. Any ideas as to what I'm doing wrong?

[Borda]

@adobles96 mind provide more information what is the implementation of your metrics = calc_aggregated_metrics(val_outputs) as it shall be a dictionary of string and number? Also, could you tell us if there is any error message or how your TB looks like?