What you'll learn

Converting raw data to features is an extremely important part of the machine learning workflow. Machine learning models are not trained on raw data, instead, they require preprocessed features that help built robust models.

In this course, Feature Sharing and Discovery Using the Databricks Feature Store, you will learn to create and use precomputed features from a centralized repository, the feature store, and the importance of feature stores and how they can help improve the machine-learning process workflow.

First, you will create and populate features in offline stores using the feature store client API, and overwrite existing features and merge new features into a store.

Next, you will learn how you can use feature lookup objects to create training sets to train machine learning models using features stored in feature tables.

Then, you will join feature store records with rows in a data frame to create training data, and log models using the feature store client and use this model to perform batch inference on your data.

Finally, you will see how you can publish your batch features to an online feature store that uses a low-latency database such as Azure Cosmos DB to store features for real-time serving. You will deploy a model to a REST endpoint and use features from the online store for real-time serving.

When you are finished with this course, you will have the skills and knowledge to use the Databricks feature store to precompute, store, and access features to train machine learning models.