TensorBoard logo


Experiment tracking

TensorBoard is a visualization toolkit to optimize and debug ML models.

Use it when

  • You want to track ML experiment metrics like loss and accuracy.
  • You want to visualize the tracked metrics and model architecture.
  • You want to view model histograms.
  • You want to log diagnostic data as images.
  • You want an integrated What-if Tool to analyze black-box classification and regression ML models.
  • You want an integrated debugger tool.

Watch out

  • What-if Tool requires TensorFlow Serving tool, and the dataset must be in a TFRecord file accessible by TensorBoard.

Example stacks

Airflow + MLflow stack