Is there a way to specify a directory for Tensorboard as it starts?

[alexgorbachev]

I need to launch Tensrboard for the directory in Google Cloud Storage (gs://bucket/directory). Any way to customize Tensorboard's directory on starting it?

[manuzhang]

It seems any directory will work if we simply pass it through. @chaoleili Any reason to read the contents of the directory first ?

[alexgorbachev]

JupyterLab UI must be passing the current directory... from the left-hand sidebar. not sure how to specify another one...

[manuzhang]

@alexgorbachev Feel free to try out https://github.com/manuzhang/jupyterlab_tensorboard/tree/remote_logdir which allows you to specify a directory

[chesterli29]

Great idea @manuzhang! It's unnecessary to check the existence of the directory. I change the start way of staring a tensorboard at this batch supportSepecifyDir. @alexgorbachev you can try this command to rebuild package:

git clone [email protected]:chaoleili/jupyterlab_tensorboard.git
git checkout supportSpecifyDir
npm install
npm run build
jupyter labextension link .

But this only supports local path, and if you want to use Google Cloud Storage, I need some time to finish this feature.

[chesterli29]

@alexgorbachev I checked jupyter_tensorboard code. it doesn't support gs or other s3 storage: https://github.com/lspvic/jupyter_tensorboard/blob/5d6fa13ecfeb27ee446ed8533c5236e21ba79b3a/jupyter_tensorboard/tensorboard_manager.py#L139, you can create an issue in jupyter_tensorboard.