Calculation Features

Overview

A Calculation is a Feature Type which allows you to define performant SQL-like operations on your source data. By avoiding the overhead of executing a Python function, Calculation features can be an efficient alternative for Feature Views that don't require the expressiveness of a python or pandas mode transformation.

Calculation Features are the recommended path for simple operations such as setting default values, simple arithmetic operations, and DATEDIFFs.

A Basic Example

transaction_analysis = RealtimeFeatureView(
    name="calculations_rtfv",
    sources=[user_metrics, req_source],
    features=[
        Calculation(name="transaction_z_score", expr="(req_source.amount - user_metrics.mean) / user_metrics.stddev"),
        Calculation(
            name="is_amount_above_threshold",
            expr="req_source.amount > user_metrics.threshold",
        ),
    ],
)

Calculations Workflow

1. Create a RealtimeFeatureView using the RealtimeFeatureView class. Realtime Feature Views that use Calculation features must be defined this way (not with a decorator).
2. Specify inputs via the sources parameter, which can include both request and entity-based sources.
3. Define each feature using the Calculation class, passing a name and SQL-style expr string.
4. Ensure the expression uses only supported functions and valid type operations. Tecton will validate syntax and types during plan/apply.
5. Apply the Feature View with tecton apply to register it for real-time serving and make it available to Feature Services.

What's Next

Once you've defined Calculation Features, consider:

• Defining a Realtime Feature View to learn more about request-time transformations.
• Exploring supported SQL expressions to understand the current capabilities of the expr syntax.
• Using this Feature View in a Feature Service for real-time feature retrieval.
• Testing with real request payloads to validate the logic and behavior of your Calculation Features.

How to Use Calculation Features

Calculation Features are defined using SQL-like expressions in the expr field of a Calculation object. Unlike python and pandas mode Realtime Feature Views which use a decorator pattern, Calculation Features are defined by directly instantiating a RealtimeFeatureView object with one or more Calculation features.

caution

Feature Views using Calculation Features can not use a transformation function.

transaction_analysis = RealtimeFeatureView(
    name="calculations_rtfv",
    sources=[user_metrics, req_source],
    features=[
        Calculation(name="transaction_z_score", expr="(req_source.amount - user_metrics.mean) / user_metrics.stddev"),
        Calculation(
            name="is_amount_above_threshold",
            expr="req_source.amount > user_metrics.threshold",
        ),
    ],
)

Type System

Calculation Features use Tecton's native type system for validating inputs and output types. When writing expressions, types are automatically inferred from your input features and operations.

Invalid type combinations will raise errors during feature validation rather than runtime:

# This will fail validation - can't add string and int
Calculation(name="invalid", expr="<string_field> + <int_field>")

SQL Expression Reference

A full list of supported SQL functionality can be found in the SDK Reference.