Set up for Vertex AI Experiments

Vertex AI Experiments is supported by the Vertex AI SDK for Python and Google Cloud console. Vertex AI Experiments requires and depends on Vertex ML Metadata. If tracking time series metrics, Vertex AI Experiments requires Vertex AI TensorBoard. (see Get started with Vertex AI TensorBoard).

Set up

• Follow steps 1-3 in Set up a project and a development environment.
• Install the Vertex AI SDK for Python client library for Python.
• Check for existence of the default Metadata Store in your project. (required)
  • To see if your project has the default Metadata Store, go to the Metadata page in the Google Cloud console.
  • If the default Metadata Store doesn't exist, it's created when
    • you run the first PipelineJob,
    • or, create your first experiment in the Vertex AI SDK for Python.
      (Optional: To configure with CMEK, see Configure your project's metadata store )
• Optional: Create a TensorBoard instance to track time series metrics.

Supported Locations

Vertex AI Experiments is available in the same locations as Vertex ML Metadata. When using Vertex AI Pipelines or Vertex AI TensorBoard, they must be in the same location as your Vertex AI experiment.

What's next

• Create an experiment

Relevant notebook tutorials

1. Compare trained and evaluated models
2. Model training with pre-built data pre-processing code
3. Compare pipeline runs
4. Autologging