The Feldera Blog
An unbounded stream of technical articles from the Feldera team
Feature Wars: SQL vs DSL for Feature Engineering
What's the best language for ML feature engineering: a specialized DSL or a general-purpose language like SQL?
How to Analyze Unbounded Time-Series Data using Bounded State
Learn how Feldera is able to run many complex SQL queries over unbounded time-series data using bounded state.
Incremental Update 9
Emit final! A quick overview of what's new in v0.29.
Taking the short path to streaming on the GPU with DBSP
Community post: DBSP is simple, general, and provides a practical mental model to the cost, both in time and in data, of computations. In this post we will put this to practice by leveraging a Python implementation of DBSP to implement incremental relational operators that show noticeable speed improvements.
Incremental Update 8
UDFs! A quick overview of what's new in v0.28.
LATENESS in streaming programs
Designing a new programming language is risky, so reusing a well-known one is usually better. Streaming systems have embraced SQL for stream processing, so we introduce a LATENESS type annotation to filter out-of-order data with a very precise definition of its behavior.
Incremental Update 7
Log Endpoints, N-Way Merging, and More! A quick overview of what's new in v0.27.
A year of magic
Announcing our rebrand, new website, and introducing Fred.
Where was Waldo (when his card was swiped)?
What was the stock price at the moment of the transaction? How far was the credit card swiped from the last known location? What was the account balance just before the transfer? These queries, common in time series analytics problems like fraud detection, can be expressed in SQL using a specialized operator: the as-of join.