oups 2025.11.3

Out-of-core pipelines over ordered data: StatefulLoop, stateful ops, and ordered Parquet Store.

Welcome to oups!

What is oups?

oups stands for Ordered Unified Processing Stack — out-of-core processing for ordered data (batch + live).

oups is a Python toolkit for building end-to-end pipelines over ordered data with the same code in offline training and live streaming/batch contexts.

It centers on StatefulLoop (loop.bind_function_state, loop.iterate, loop.buffer), which binds and persists function/object state, orchestrates chunked iteration, and buffers DataFrames under a memory cap with flush-on-limit or last-iteration semantics. Complementing the loop, stateful_ops provides vectorized, chunk-friendly primitives like AsofMerger for multi-DataFrame as-of joins (with optional windows of previous values) and SegmentedAggregator (planned) for streamed segmentation and aggregation. The store package manages ordered Parquet datasets via schema-driven keys (@toplevel), supports incremental updates (store[key].write(...)) and duplicate handling, and offers synchronized iteration across datasets via store.iter_intersections(...) with optional warm-up (n_prev).

Together these pieces enable out-of-core processing with resumability, and deterministic buffering. The design favors explicit, minimal APIs and reproducible results, aligning offline feature generation with online serving.

Links

• 📖 Documentation - Guides and API reference
• 📋 Changelog - Release notes and version history