Taking a slight detour from my ongoing MLOps project to delve a bit deeper into Hopsworks.
These are my random notes in no particular order, since starting to work on the MLOps project:
The Purpose of Feature Stores: At heart, feature stores are dedicated spaces for storing features essential for machine learning models. These features have already undergone the process of feature engineering.
Exploring External Tables in Hopsworks: Hopsworks offers support for external tables such as JDBC-enabled source, Snowflake, Data Lake, Redshift, BigQuery, S3, and others as feature groups. This is something I've yet to try, but I'm keen to explore. Their JDBC connector is not great yet.
Feature Views: Hopsworks 'Feature Views' are essentially a composite of features drawn from various feature groups. I haven't had the chance to use feature views yet.
Easy Use with Python: Hopsworks has user-friendly interfaces, especially Python.
Feature Groups Versioning: Hopsworks allows versioning of feature groups. This is an advantage over using just a simple SQL table. It works for having a history track of your features and allows better management and control over ML model's data.
Silverboy
Top comments (0)