TensorBoards are visualizations of the machine learning model training process. They work by continually reading log files while a model is training.

As you train a model using a notebook, you can connect your model log files and data set to a TensorBoard to see its progress in graph form as it trains.

Screenshot of the TensorBoards page upon first launch


Before you create a TensorBoard, you need to specify the directory where your log files will be stored.

In a Notebook, change the log_dir variable to point at a data store. This could be a new Volume on ML Workbench or on external cloud storage.

When you create a TensorBoard, point it to that log file directory so that it can interpret the log files and display the model training progress visually.

Create a TensorBoard

new tensorboard
  1. Click on the + New TensorBoard button in the upper right corner of the Tensorboards tab.

  2. Enter the name.

  3. Choose the location of your data either as an Object Store link or a PVC and mount path. For a TigerGraph Cloud solution, the log files will be in a PVC.

After a few moments of initialization, the new TensorBoard appears in the list. Click Connect to open a new tab to view the TensorBoard.