Predictive ML

Time-Window Aggregations

Optimized for the most common and computationally demanding ML features

Gain performance and efficiency with the aggregation engine

Tecton’s aggregation engine enables efficient data pipeline construction at scale. By splitting compute across write and read time and compacting large amounts of event data into compressed data points, it reduces costs for computationally expensive features. This allows for complex aggregations that were previously impractical or cost-prohibitive, extracting more value from data while optimizing resources.

Aggregate extremely fresh

For time-sensitive applications like fraud detection, Tecton’s time window aggregations allow features to have extreme freshness, enabling up-to-the-second aggregations. Unlike most streaming engines, time window aggregations with Tecton are kept up-to-date even if no new events arrive. This capability enables models to make decisions with more current data, providing an advantage in real-time scenarios.

Support an extensive range of time windows

Tecton’s time window aggregations handle data of any scale with a single, unified API. Whether you’re processing minute-level snapshots, aggregations over sliding time windows or saw-tooth time windows, or even lifetime aggregations across massive datasets, Tecton efficiently manages your data. Capture trends and patterns at any scale, enabling more accurate and relevant features for ML models regardless of data volume or velocity.

Combine aggregations with custom logic

Tecton goes beyond simple statistical functions, allowing you to combine aggregations with row-level Python operations. Choose from a variety of built-in aggregation functions, or create custom aggregations to satisfy unique requirements.

Unify stream and batch aggregations

Define aggregations once, and apply them automatically to both historical backfills and new streaming data. This unified approach simplifies your data architecture, reduces code duplication, and ensures consistency across the entire data processing pipeline. Tecton also automates point-in-time correct backfilling, making it easy to generate historically accurate features for training data while maintaining consistency between offline and online environments.