Featureform - The Open-Source Virtual Feature Store (original) (raw)

Revolutionizing How ML Teams Work with Data

Facilitate Deployment

Define your features once with Featureform, and we’ll orchestrate your transformation pipelines for both training and inference, across batch and streaming.

Enhance Collaboration

All transformations and features are searchable, re-usable, and extensible. The days of sending notebooks and datasets over slack is over.

Organize Experimentation

All changes in your feature pipelines are automatically captured, versioned, and saved. They can be re-used across different notebooks so all your features are unified in a single resource repository.

Increase Reliability

All your production feature pipelines are actively monitored for job failures, data drift, and more, allowing you to preemptively catch and resolve failures.

Preserve Compliance

With built-in role based access control, audit logs, and dynamic serving tools along with native integrations into your data catalogs and identity providers, Featureform preserves compliance in the ML stack.

Define, Manage, and Deploy Your ML Features

DECLARATIVE API

DASHBOARD

INFERENCE

TRAINING

RESOURCE REPOSITORY

COLLABORATION

MONITORING & ALERTING

GOVERNANCE

ORCHESTRATION

INFRASTRUCTURE PROVIDERS

DATABASE

Featureform in Action

Define Data Sets, Features, and Training Sets

Featureform makes data management for machine learning easy. Our Python API enables data scientists to define datasets, transformations, features, and training sets. Drawing inspiration from Terraform, it promotes a declarative approach where every data pipeline, from raw input to processed feature, is transparent and traceable. It operates seamlessly on your existing infrastructure including Databricks and Snowflake. Featureform fosters a robust, versioned, and lineage-rich data workflow for machine learning.

Monitor and Alerts for Failures and Data Drift

Featureform serves as a collaborative repository, enabling easy access and sharing of datasets and features through APIs and a user-friendly dashboard. The platform offers robust search, monitoring, and lineage tools, allowing users to efficiently discover and understand the characteristics of datasets and features. It ensures immutability of versions, fostering consistent, reliable collaboration and model building across data science teams.

Streaming Data, On-Demand Features, and Real-Time Serving

Featureform simplifies the process for data scientists to define real-time, bath, or on-demand features, handling technical complexities like backfill and materialization into stores like DynamoDB or Redis. Its backfill capabilities ensure historical integrity of datasets, while point-in-time correctness guarantees accurate model training and predictions. This empowers teams to deploy sophisticated features with minimal effort, focusing on innovation rather than infrastructure.

@ff entity class User:
  last_purchase = ff.FeatureStream(offline_store=snowflake,
online_store=redis, type=ff.Float32)

client write_feature(User. last_purchase, (user_id,
new_value) )

@ff .ondemand_feature()
def ondemand_percent(client, params, entities) :
  import featureform as ff
  return params[ "TransactionAmount"] /
client features(I("balance", ff.get_run())],
entities=entities)

Feature Governance and Compliance

Featureform integrates with identity providers like Okta and various data catalogs for robust governance and access control, synchronizing with these systems to maintain a single source of truth. It offers granular access controls for precise management of data and feature access, aligning with an organization's security protocols. This approach supports a seamlessly compliant machine learning workflow, knowing that compliance, privacy, and proper access are all in check with your other systems.

A Feature Platform Built for Enterprise

Access Control & Governance

Ensure data integrity and regulatory compliance effortlessly, while seamlessly tracking data lineage, preserving data quality, and proactively mitigating bias and privacy risks.

Streaming

Keep models responsive to changing trends, enhancing their accuracy and relevance. With streaming data, you can swiftly adapt to emerging opportunities, detect anomalies, and provide personalized experiences.

On-Prem Deployments

Our deployment model ensures that our solution is seamlessly integrated into your cloud or on-premise environment, giving you complete control.

Custom Integrations

Featureform’s plug-in architecture will seamlessly integrate into your existing tooling, including data infrastructure, identify providers, and monitoring tools. If you have a custom provider or configuration that isn’t in our open-source product, we can build it for you.

Join Our Community on Slack

Network with people in the MLOps community and get live support from our team.

Explore Our Documentation

Find more details on what the FeatureForm workflow can do for you, as well as what sets it apart from rival feature store platforms.

Continue Learning

Our Resource Hub has the educational tools to equip you with the most recent innovations in MLOps.

Ready to get started?

See what a virtual feature store means for your organization.