A scalable, unified data and AI engineering platform for enterprise

Important Links: Slack & Discussions. Docs.

What is Feathr?

Feathr is a data and AI engineering platform that is widely used in production at LinkedIn for many years and was open sourced in 2022. It is currently a project under LF AI & Data Foundation.

Read our announcement on Open Sourcing Feathr and Feathr on Azure, as well as the announcement from LF AI & Data Foundation.

Feathr lets you:

• Define data and feature transformations based on raw data sources (batch and streaming) using Pythonic APIs.
• Register transformations by names and get transformed data(features) for various use cases including AI modeling, compliance, go-to-market and more.
• Share transformations and data(features) across team and company.

Feathr is particularly useful in AI modeling where it automatically computes your feature transformations and joins them to your training data, using point-in-time-correct semantics to avoid data leakage, and supports materializing and deploying your features for use online in production.

🌟 Feathr Highlights

• Native cloud integration with simplified and scalable architecture.
• Battle tested in production for more than 6 years: LinkedIn has been using Feathr in production for over 6 years and backed by a dedicated team.
• Scalable with built-in optimizations: Feathr can process billions of rows and PB scale data with built-in optimizations such as bloom filters and salted joins.
• Rich transformation APIs including time-based aggregations, sliding window joins, look-up features, all with point-in-time correctness for AI.
• Pythonic APIs and highly customizable user-defined functions (UDFs) with native PySpark and Spark SQL support to lower the learning curve for all data scientists.
• Unified data transformation API works in offline batch, streaming, and online environments.
• Feathr’s built-in registry makes named transformations and data/feature reuse a breeze.

🏃 Getting Started with Feathr - Feathr Sandbox

The easiest way to try out Feathr is to use the Feathr Sandbox which is a self-contained container with most of Feathr’s capabilities and you should be productive in 5 minutes. To use it, simply run this command:

# 80: Feathr UI, 8888: Jupyter, 7080: Interpret
docker run -it --rm -p 8888:8888 -p 8081:80 -p 7080:7080 -e GRANT_SUDO=yes feathrfeaturestore/feathr-sandbox:releases-v1.0.0

And you can view Feathr quickstart jupyter notebook:

http://localhost:8888/lab/workspaces/auto-w/tree/local_quickstart_notebook.ipynb

After running the notebook, all the features will be registered in the UI, and you can visit the Feathr UI at:

http://localhost:8081

🛠️ Install Feathr Client Locally

If you want to install Feathr client in a python environment, use this:

pip install feathr

Or use the latest code from GitHub:

pip install git+https://github.com/feathr-ai/feathr.git#subdirectory=feathr_project

☁️ Running Feathr on Cloud for Production

Feathr has native integrations with Databricks and Azure Synapse:

Follow the Feathr ARM deployment guide to run Feathr on Azure. This allows you to quickly get started with automated deployment using Azure Resource Manager template.

If you want to set up everything manually, you can checkout the Feathr CLI deployment guide to run Feathr on Azure. This allows you to understand what is going on and set up one resource at a time.

• Please read the Quick Start Guide for Feathr on Databricks to run Feathr with Databricks.
• Please read the Quick Start Guide for Feathr on Azure Synapse to run Feathr with Azure Synapse.

📓 Documentation

• For more details on Feathr, read our documentation.
• For Python API references, read the Python API Reference.
• For technical talks on Feathr, see the slides here and here. The recording is here.

🧪 Samples

Name	Description	Platform
NYC Taxi Demo	Quickstart notebook that showcases how to define, materialize, and register features with NYC taxi-fare prediction sample data.	Azure Synapse, Databricks, Local Spark
Databricks Quickstart NYC Taxi Demo	Quickstart Databricks notebook with NYC taxi-fare prediction sample data.	Databricks
Feature Embedding	Feathr UDF example showing how to define and use feature embedding with a pre-trained Transformer model and hotel review sample data.	Databricks
Fraud Detection Demo	An example to demonstrate Feature Store using multiple data sources such as user account and transaction data.	Azure Synapse, Databricks, Local Spark
Product Recommendation Demo	Feathr Feature Store example notebook with a product recommendation scenario	Azure Synapse, Databricks, Local Spark

🔡 Feathr Highlighted Capabilities

Please read Feathr Full Capabilities for more examples. Below are a few selected ones:

Feathr UI

Feathr provides an intuitive UI so you can search and explore all the available features and their corresponding lineages.

You can use Feathr UI to search features, identify data sources, track feature lineages and manage access controls. Check out the latest live demo here to see what Feathr UI can do for you. Use one of following accounts when you are prompted to login:

• A work or school organization account, includes Office 365 subscribers.
• Microsoft personal account, this means an account can access to Skype, Outlook.com, OneDrive, and Xbox LIVE.

For more information on the Feathr UI and the registry behind it, please refer to Feathr Feature Registry

Rich UDF Support

Feathr has highly customizable UDFs with native PySpark and Spark SQL integration to lower learning curve for data scientists:

def add_new_dropoff_and_fare_amount_column(df: DataFrame):
    df = df.withColumn("f_day_of_week", dayofweek("lpep_dropoff_datetime"))
    df = df.withColumn("fare_amount_cents", df.fare_amount.cast('double') * 100)
    return df

batch_source = HdfsSource(name="nycTaxiBatchSource",
                        path="abfss://feathrazuretest3fs@feathrazuretest3storage.dfs.core.windows.net/demo_data/green_tripdata_2020-04.csv",
                        preprocessing=add_new_dropoff_and_fare_amount_column,
                        event_timestamp_column="new_lpep_dropoff_datetime",
                        timestamp_format="yyyy-MM-dd HH:mm:ss")

Defining Window Aggregation Features with Point-in-time correctness

agg_features = [Feature(name="f_location_avg_fare",
                        key=location_id,                          # Query/join key of the feature(group)
                        feature_type=FLOAT,
                        transform=WindowAggTransformation(        # Window Aggregation transformation
                            agg_expr="cast_float(fare_amount)",
                            agg_func="AVG",                       # Apply average aggregation over the window
                            window="90d")),                       # Over a 90-day window
                ]

agg_anchor = FeatureAnchor(name="aggregationFeatures",
                           source=batch_source,
                           features=agg_features)

Define Features on Top of Other Features - Derived Features

# Compute a new feature(a.k.a. derived feature) on top of an existing feature
derived_feature = DerivedFeature(name="f_trip_time_distance",
                                 feature_type=FLOAT,
                                 key=trip_key,
                                 input_features=[f_trip_distance, f_trip_time_duration],
                                 transform="f_trip_distance * f_trip_time_duration")

# Another example to compute embedding similarity
user_embedding = Feature(name="user_embedding", feature_type=DENSE_VECTOR, key=user_key)
item_embedding = Feature(name="item_embedding", feature_type=DENSE_VECTOR, key=item_key)

user_item_similarity = DerivedFeature(name="user_item_similarity",
                                      feature_type=FLOAT,
                                      key=[user_key, item_key],
                                      input_features=[user_embedding, item_embedding],
                                      transform="cosine_similarity(user_embedding, item_embedding)")

Define Streaming Features

Read the Streaming Source Ingestion Guide for more details.

Point in Time Joins

Read Point-in-time Correctness and Point-in-time Join in Feathr for more details.

Running Feathr Examples

Follow the quick start Jupyter Notebook to try it out. There is also a companion quick start guide containing a bit more explanation on the notebook.

🗣️ Tech Talks on Feathr

• Introduction to Feathr - Beginner’s guide
• Document Intelligence using Azure Feature Store (Feathr) and SynapseML
• Notebook tutorial: Build a Product Recommendation Machine Learning Model with Feathr Feature Store
• Feathr talk in Feature Store Summit

⚙️ Cloud Integrations and Architecture

Feathr component	Cloud Integrations
Offline store – Object Store	Azure Blob Storage, Azure ADLS Gen2, AWS S3
Offline store – SQL	Azure SQL DB, Azure Synapse Dedicated SQL Pools, Azure SQL in VM, Snowflake
Streaming Source	Kafka, EventHub
Online store	Redis, Azure Cosmos DB
Feature Registry and Governance	Azure Purview, ANSI SQL such as Azure SQL Server
Compute Engine	Azure Synapse Spark Pools, Databricks
Machine Learning Platform	Azure Machine Learning, Jupyter Notebook, Databricks Notebook
File Format	Parquet, ORC, Avro, JSON, Delta Lake, CSV
Credentials	Azure Key Vault

🚀 Roadmap

• More Feathr online client libraries such as Java
• Support feature versioning
• Support feature monitoring

👨‍👨‍👦‍👦 Community Guidelines

Build for the community and build by the community. Check out Community Guidelines.

📢 Slack Channel

Join our Slack channel for questions and discussions (or click the invitation link).