Help for Tensorboard logging data to Grafana dashboards?d

Hello, my goal is to be able to make Grafana dashboards using logging data that is currently in Tensorboard format.

I’d love some help and/or recommendations for what data formats to convert the Tensorboard Logging, and how to use that data as a datasource for Grafana.

My requirements are that everything must be local, so I cannot store data in a cloud, and that I must be able to implement this in soft-real-time, so more information will be logged and I’d like that updated data to be reflected on Grafana.

Right now I am considering converting the Tensorboard data to CSV or pandas Dataframes, and hopefully sending that to Prometheus to use as a datasource for Grafana.

If you have any advice, recommendations, or a way to make this pipeline simpler, it’d be extremely appreciated. Thank you.


Please provide details about this data format

Or provide a sample data of what it looks like

Or is there a rest api provided by tensorboard?

This is how the data is being logged, using pytorch lightning tensorboardlogger: TensorBoardLogger — PyTorch Lightning 2.0.2 documentation

Here is more about the data format: TensorBoard  |  TensorFlow

1 Like

Cool, but still waiting for you to provide sample data

Is it jaon, csv or …?

Unfortunately, I can’t share the data. The data is being logged as a tensorboard object file which doesn’t seem legible to humans, but I think it can be converted to csv or pandas.dataframe >python - How do you read Tensorboard files programmatically? - Stack Overflow
… I am hoping to find help or recommendations of how to get from tensorboard to grafana.

Cool. Do you know how to use python to convert that data to csv or json,? If so please post thr structure of that csv or json data here

Unfortunately, I do not at the moment. I am posting on the tensorflow public help forum as well for help. Is there no avenue to send Tensorboard data directly to grafana or prometheus?

1 Like