Trainer.add_argparse_args(parser) break the default Tensorboard hparams logging.

[lkhphuc]

🐛 Bug

Trainer.add_argparse_args(parser) break the default Tensorboard hparams logging.

To Reproduce

Steps to reproduce the behavior:

I pretty much just put together the sample codes in the Hyperparameters section in the docs and it's throw the error.

Code sample

class LitMNIST(pl.LightningModule):
  def __init__(self, hparams):
    super(LitMNIST, self).__init__()
    self.hparams = hparams

    self.layer_1 = torch.nn.Linear(28 * 28, hparams.layer_1_dim)

  def forward(self, x):
    return self.layer_1(x)

  def train_dataloader(self):
    return DataLoader(mydata(), batch_size=self.hparams.batch_size)

  def configure_optimizers(self):
    return Adam(self.parameters(), lr=self.hparams.learning_rate)


def main(args):
    model = LitMNIST(args)
    trainer = pl.Trainer()
    trainer.fit(model)


if __name__ == "__main__":
    parser = ArgumentParser()

    # parametrize the network
    parser.add_argument('--layer_1_dim', type=int, default=128)
    parser.add_argument('--learning_rate', type=float, default=1e-3)

    # add all the available options to the trainer
    parser = pl.Trainer.add_argparse_args(parser)

    args = parser.parse_args()
    main(args)

Traceback (most recent call last):
  File "tmp.py", line 56, in <module>
    main(args)
  File "tmp.py", line 40, in main
    trainer.fit(model)
  File "/Users/phuc/miniconda3/envs/thinc/lib/python3.7/site-packages/pytorch_ligh
tning/trainer/trainer.py", line 630, in fit
    self.run_pretrain_routine(model)
  File "/Users/phuc/miniconda3/envs/thinc/lib/python3.7/site-packages/pytorch_ligh
tning/trainer/trainer.py", line 748, in run_pretrain_routine
    self.logger.log_hyperparams(ref_model.hparams)
  File "/Users/phuc/miniconda3/envs/thinc/lib/python3.7/site-packages/pytorch_ligh
tning/loggers/base.py", line 18, in wrapped_fn
    fn(self, *args, **kwargs)
  File "/Users/phuc/miniconda3/envs/thinc/lib/python3.7/site-packages/pytorch_ligh
tning/loggers/tensorboard.py", line 113, in log_hyperparams
    exp, ssi, sei = hparams(params, {})
  File "/Users/phuc/miniconda3/envs/thinc/lib/python3.7/site-packages/torch/utils/
tensorboard/summary.py", line 156, in hparams
    raise ValueError('value should be one of int, float, str, bool, or torch.Tenso
r')
ValueError: value should be one of int, float, str, bool, or torch.Tensor

The value it fails at is key callback with value [].

Expected behavior

Trainer.add_argparse_args(parser) should not create trouble for tensorboard hparams logging.

Environment

• PyTorch Version (e.g., 1.0): 1.3.1
• OS (e.g., Linux): Linux
• How you installed PyTorch (conda, pip, source): pip
• Python version: 3.7.5

[awaelchli]

Hi, not sure but it looks like this got fixed on master recently. See #1130.
Could you try installing from master and try again?

pip install git+https://github.com/PytorchLightning/pytorch-lightning.git@master --upgrade

[lkhphuc]

Yep that fixes it. Thanks.