Otávio Aidar

Dear Nixtla Developers,

I am writing to report a persistent issue I've been encountering while using the NeuralForecast library. Every time I attempt to run any code, regardless of the model architecture or data used, the execution consistently results in an error related to the logging directory.

I have tried various troubleshooting steps, including:

Specifying a new logging directory using the

default_root_dir

argument in the

pl.Trainer

class.

Configuring the TensorBoard logger to use the new directory.

Disabling the default PyTorch Lightning logger with

logger=False

.

Uninstalling and reinstalling TensorBoard.

Setting the

TENSORBOARD_LOGDIR

environment variable.

Restarting the Jupyter Notebook kernel.

Despite these efforts, the error persists. The error message I consistently receive is:

```

FailedPreconditionError: C:\Users\Otávio\lightning_logs is not a directory

This suggests a conflict with the default logging directory used by PyTorch Lightning or TensorBoard. I have confirmed that the specified new logging directory exists and that I have write permissions.

I am using the following library versions:

NeuralForecast: neuralforecast==1.7.3

PyTorch Lightning: pytorch-lightning==2.3.0

PyTorch: torch==2.3.0

TensorFlow: tensorflow==2.16.1

I am running my code on Jupyter Notebook .

I would greatly appreciate any assistance you can provide in resolving this issue. Please let me know if you require any further information or code snippets to help diagnose the problem.

Thank you for your time and support.

Sincerely,

Otávio.