On September 15, 2022, Verily's Dr. Amy Unruh presented a virtual webinar that reviewed different ways you can use Terra for machine learning (ML), with a particular focus on Google Cloud (GC). She used the public Terra workspace for ML on Terra, which you can try yourself. The workspace contains Jupyter Notebooks that use a Keras model to run an image classification task on sample tissue.
Materials
- The public workspace is available at https://app.terra.bio/#workspaces/verily-terra-solutions/ml-on-terra.
- The slides are at https://bit.ly/ml-terra-sept15.
- Terra registration instructions are here.
-
Jupyter Notebook code is available GitHub: https://github.com/
DataBiosphere/terra-examples/ tree/main/ml_notebooks/image_ classification - A recording of the webinar is available here.
Agenda
In this session, participants learned how to:
- Run training jobs directly in a workspace notebook
- Access Vertex AI, BQML (BigQuery ML), and other GC services from a Terra notebook
- Use Vertex AI for model training and serving (online prediction)
- Use Vertex AI to do hyperparameter tuning
- Use the Vertex Experiments API and Managed TensorBoard service to log and monitor your experiments and training progress
- Build and run ML workflows using Vertex AI Pipelines— using both prebuilt components (task definitions) and your own custom components
- Train BQML models from BigQuery data and use them for prediction