Set trainer_config = {'logger':none} not work

[Aaron1993]

when I trained the NeuralProphet model, I won't save the log file named 'lightning_logs', so I set the trainer_config = {'logger': None/False}, and it still saved the log file. How can I resolve this problem?

[ourownstory]

Hi @Aaron1993 Can you please provide some details on what behavior you are experiencing vs expecting?

[samialitop3]

I am also facing the same issue when trying to run the NeuralProphet model inside AWS Lambda function.

AWS Lambda function do not allow writing files except for the /tmp directory, so I have tried to set the trainer_config default_root_dir to the /tmp directory but it did not work.

Here is how I configured the NeuralProphet model

        m = NeuralProphet(
            normalize="off",
            daily_seasonality=True,
            trainer_config={"default_root_dir": "/tmp"},
        )

and I get this error when I access the Lambda function logs

OSError: [Errno 30] Read-only file system: '/var/task/lightning_logs'

[samialitop3]

What I expect when I set trainer_config = {'logger': None/False} is to disable the pytorch_lightning logger or setting the default_root_dir to generate the lightning_logs in the specified directory.

[WuttsGood]

I'm also running into this issue. Is there a way to create the lightning_logs folder in a different directory?

[jennank]

Bump on this, I am also having this issue. Is there a way to turn off the logging to also get around this?

[parksjr5]

Bumping, thanks!

[samialitop3]

Bumping on this!

[baylenspringer-readysignal]

I am experiencing this issue as well. Bumping, thanks!

[samialitop3]

Bumping on this!

[samialitop3]

@ourownstory Can you please check this for any possible fixes?

[sugyeong-jo]

I would like to solve this problem too. I would like to use this package multi-threaded, but this folder creation seems to be preventing it.

[csaid]

I'm experiencing this when running NeuralProphet in Snowflake's Snowpark, which will only let me write to the /tmp folder. The error I'm getting is:

OSError: [Errno 30] Read-only file system: '/.lr_find_494580fa-8958-4d83-ae9c-ae202c99ae01.ckpt'

Snowpark lists some ways for common libraries to write specifically to the /tmp folder. https://docs.snowflake.com/en/user-guide/ui-snowsight/notebooks-troubleshoot#read-only-file-system-issue

Could we have a way to do this in NeuralProphet?

[crocoroc0]

For those running NeuralProphet in read-only environments (like AWS Lambda) and encountering ReadOnlyFileSystemError:

The core workaround is to temporarily switch the Current Working Directory (CWD) to a writable path (e.g., /tmp/some_dir) just before initialising NeuralProphet, and then switch it back immediately after.

You also need to set default_root_dir. This is what worked for me:

temp_np_workdir = "/tmp/neuralprophet_init_workdir" # Writable temporary directory

try:
    os.makedirs(temp_np_workdir, exist_ok=True)
    os.chdir(temp_np_workdir)

    # Configure PyTorch Lightning to use /tmp/ for its default outputs
    trainer_config_for_tmp = {
        "default_root_dir": "/tmp/ptl_trainer_output/"
    }

    m = NeuralProphet(
        trainer_config=trainer_config_for_tmp
        # ... other NeuralProphet parameters
    )

finally:
    os.chdir(original_cwd) 

[csaid]

Thank you @crocoroc0 ! That worked.