omuratgultekin / feast_feature_store Goto Github PK

This is the complete feature store that was built for Feature Storage for ML with Feast: Part 1 – Building a Local Feature Store Infrastructure for Training and Prediction.

The feature store by itself is located in breast_cancer. The Jupyter notebook and Python scripts in the root directory of the repo are intended to help you prepare data, fetch features, train a model, and do inference.

Note that the data files in breast_cancer/data will very likely be outdated by the time you see this repository. After cloning the repo and before using the code, make sure to:

1. Run the Jupyter notebook feast_data_preparation.ipynb in the repo's root path to generate a toy dataset for the feature store.
2. Move the five generated files - data_df1.parquet, data_df2.parquet, data_df3.parquet, data_df4.parquet, and target_df.parquet - to breast_cancer/data.

Additionally, change the paths provided to to the class FileSource in breast_cancer/definitions.py

To find out more about how to use Feast, read official Feast docs and also Feature Storage for ML with Feast: Part 1 – Building a Local Feature Store Infrastructure for Training and Prediction.