FeatureByte: A Comprehensive Workflow for Feature Engineering

FeatureByte streamlines the end-to-end feature engineering process, from feature creation to deployment and lifecycle management.

Ready to get started? Explore each step and its corresponding tutorials to unlock the full potential of FeatureByte.

---

Step 1: Define Your Data Model

1. Create a Catalog: Organize your data into a clear and accessible structure.
2. Register Tables: Classify tables as event, item, time series, dimension, or slowly changing dimension.
3. Define entities: Identify key entities within your data and tag relevant columns.
4. Enhance Data Quality: Set default cleaning operations to ensure data accuracy.

Tutorials

• Credit Default UI Tutorials: Create Catalog, Register Tables, Register Entities, Update Descriptions and Tag Semantics, Set Default Cleaning Operations.

• Credit Default SDK Tutorials: Create Catalog, Register Tables, Register Entities, Update Descriptions, Set Default Cleaning Operations.

• Grocery UI Tutorials: Create Catalog, Register Tables, Register Entities, Update Descriptions and Tag Semantics, Set Default Cleaning Operations.

• Grocery SDK Tutorials: Create Catalog, Register Tables, Register Entities, Update Descriptions, Set Default Cleaning Operations.

---

Step 2: Formulate Your Use Case

1. Identify the Primary Entity: Determine the central focus of your use case.
2. Define the Target: Specify the outcome you aim to predict or classify.
3. Establish the Context: Define the scope and circumstances in which features are expected to be served.
4. Create Observation Sets: Define the specific data points to be used for modeling.

Tutorials

• Credit Default UI Tutorials: Formulate Use Case, Create Observation Tables.

• Credit Default SDK Tutorials: Formulate Use Case, Create Observation Tables.

• Grocery UI Tutorials: Formulate Use Case, Create Observation Tables.

• Grocery SDK Tutorials: Formulate Use Case, Create Observation Tables.

---

Step 3: Create Features

Automated Feature Creation:

• Use Feature Ideation to generate relevant feature suggestions.
• Evaluate features using EDA and select the most promising ones with various feature selection strategies.

Tutorials

• Credit Default UI Tutorials: Ideate Features, Refine Ideation

• Grocery UI Tutorials: Ideate Features, Refine Ideation

Manual Feature Creation:

• Leverage the Python SDK to create custom features.

Tutorials

• Credit Default SDK Tutorials: Create Lookup features, Create Window Aggregates from Event Table, Create Features from SCD, Create Calendar Window Aggregates from Time Series,

• Grocery SDK Tutorials: Create lookup feature, Create window aggregate features, Derive features from other features, Derive similarity features from bucketing, Use embeddings,

---

Step 4: Experiment and Iterate

1. Build feature lists: Combine selected features into cohesive sets using Feature Ideation, the Feature List Builder, or the SDK.
2. Generate Historical Feature Data: Prepare historical data for model training and evaluation.
3. Train and Test Models: Iterate on model development and hyperparameter tuning.
4. Share Promising features: Collaborate with team members and share valuable insights.

Tutorials

• Credit Default UI Tutorials: Create New Feature List, Compute Feature Table.

• Credit Default SDK Tutorials: Create Feature Lists, Compute Historical Feature Values.

• Grocery UI Tutorials: Create New Feature List, Compute Feature Table.

• Grocery SDK Tutorials: Create Feature Lists, Compute Historical Feature Values.

---

Step 5: Deploy and Serve Features

1. Mark features as production-ready: Certify features for deployment.
2. Deploy Features and Enable Serving: Make features available for real-time or batch inference.

In catalogs with Approval Flow enabled, deploying features involves:

• Verifying feature compliance with default cleaning operations and feature job settings.
• Checking the status of source tables.
• Backtesting feature job settings to ensure no future training-serving inconsistencies.
• Sharing the feature definition file for review and approval.

This comprehensive process ensures governance and reduces the risk of errors in production.

Tutorials

• Credit Default UI Tutorials: Deploy and Serve

• Credit Default SDK Tutorials: Deploy and Serve

• Grocery UI Tutorials: Deploy and Serve

• Grocery SDK Tutorials: Deploy and Serve

---

Step 6: Manage the Feature Life Cycle

1. Update Feature Job Settings and Cleaning Operations: Adjust configurations when data quality or availability changes.
2. Create New Feature Versions: Produce updated versions reflecting the latest settings and mark them as default.
3. Monitor Feature Job Status: Regularly review feature job status to track performance and ensure smooth operations.

In catalogs with Approval Flow enabled, changes in table metadata trigger a review process. This ensures:

• New versions of features and feature lists address any data-related issues.
• New deployments use the latest, approved configurations.
• Clear documentation of all changes for compliance and reproducibility.

Tutorials

• Grocery UI Tutorials: Manage Life Cycle
• SDK Tutorials: Manage Life Cycle