View Hparam in log

implementation help
1

Please tell me how I can save, display and compare the launch parameters of the model, in some dashboard like tensorboard

2

You shall be able to see all hyper params in your TensorBoard if you have saved them in your model

class MyModel(LightningModule):
  def __init__(self, my_arg):
    super().__init__()
    ...
    self. save_hyperparameters()

for more information, see ref in docs
evetualy you can have extra call to save your hyper-parameters from Trainer like
trainer.logger.log_hyperparams(model.hparams_initial)

1 Like
3

tnx!
it’s great work!

hp_metric you know how add in end train, for best values of val set?

4

You can use for example
trainer.logger.log_metric('hp_metric', trainer.callback_metrics['loss])
at the end of training, for example at on_fit_end() or i do it on_epoch_end(), so if training brake i got last epoch value.

5

if you are using a checkpointing system, and the hp_metric is what you monitor, you can call checkpoint attribute best_model_score, see the link to the code