TensorBoard is a powerful visualization tool when building machine learning models. However it can quickly become quite unwieldy as the number of model versions and runs increases.
TensorBoard Projects provides a UI to manage Tensorboard runs and allows you to easily:
- Visualize runs for a given model
- Archive Runs
- Delete Runs
- Add metadata to individual runss
- Start TensorBoard dashboard for a subset of model runs
- Write documentation for a model
As the project is still very much in development, please report any issues or features you would like to see added !
TensorBoard projects is not an experiment tracking solution, it simply allows you to better manage TensorBoard runs.
If you are looking for a fully fledgeed experiment tracking solution, you can look into:
- ClearML
- Comet
- MLFlow
- Weights and Biases and others
Tensorboard Projects can be installed from PyPI using pip install tensorboard-projects.
Once Tensorboard Projects is installed, you can start the UI using:
tensorboard-projects ui
In order to assist with running this on a remote machine, the following arguments are supported:
--backend-store-uri: Storage location of metadata, defaults to~/.tensorboard_projects--ip: Host for the API, use 0.0.0.0 to access from a remote machine--port: Port to run UI on remote machien