Hamilton: an open source framework for feature transform code in Python
summary
Hamilton is an open source Python micro framework that models feature transforms as named functions with explicit inputs, a driver that scrapes modules to build a DAG, and runtime checks for data quality. It runs with pandas and any Python object, embeds into Airflow or Feast, and supports online/async use. At Stitch Fix it curates 4,000 feature transform definitions, makes every feature unit testable and documented, and cuts monthly update work to 4x faster by enforcing a declarative data flow.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:37What is Hamilton
- 4:14Why was it created
- 7:48Hamilton at Stitch Fix
- 8:51How to use Hamilton
- 11:59Problems with feature engineering
- 14:51Functions
- 17:51Summary
- 18:33Questions
related talks
31mHamilton:Natively bringing SWE best practices to python data transformations
41mScalable Feature Engineering with Hamilton
30mBridging Classic ML Pipelines with the World of LLMs
40mEnabling simple ML ops at Stitch Fix
36mDeclarative Feature Engineering
38mFeatran, Featran77 - Generic Feature Transformer for Data Pipelines
32mA feature Store for Machine Learning
26mIn house to open source Stitching the past to the futu...