RINSE

Overview

RINSE, short for RINSE INteractive Series Explorer, is a prototype exploration tool for large collections of data series. It is built around adaptive indexing so users can start querying quickly instead of waiting for full indexing pipelines to complete.

The motivating problem is broad. Similarity search over data series matters in meteorology, chemistry, finance, smart cities, marketing, and many other domains where analysts need to identify patterns, detect anomalies, and compare evolving signals.

Traditional indexing approaches can take days to prepare multi-terabyte collections. That delay becomes a show-stopper when users need immediate access, exploratory analysis, or only a modest number of queries. RINSE addresses this by pairing interactive querying with an index that grows only where the workload actually needs it.

Why Data Series?

Large time series collections appear across science and industry. Similarity search is one of the core primitives for understanding those datasets.

Why RINSE?

RINSE is designed for exploration-first workflows where waiting for a fully materialized index is too expensive or simply unnecessary.

Highlights

Adaptive indexing turns interactive exploration into a first-class systems goal.

ADS

Adaptive Indexing

Build only the necessary tree structure before queries arrive

Seconds

Interactive Queries

Explore large collections without multi-day startup cost

Exact + Approx

Search Modes

Support different tradeoffs during interactive exploration

Research

RINSE is built on adaptive indexing for data series and demonstrates how exploration can begin long before a conventional index would be ready.

ADS / ADS+

Adaptive Data Series indexing builds a lightweight tree skeleton first and materializes raw data only when queries need it.

Interactive Queries

Users can draw queries, select examples, or generate random probes from the web interface.

Three Access Paths

RINSE exposes serial scan, iSAX 2.0, and adaptive ADS+ to compare behavior on the same workload.

Progressive Ingestion

Only the data relevant to the incoming workload is pulled into the adaptive index over time.

During indexing, ADS performs only a few basic steps. It creates the skeleton of a tree that stores condensed information about the input series. Leaves initially contain symbolic representations but no raw series, so the system can become queryable much faster than up-front indexing designs.

When a query arrives, RINSE converts it, traverses the index, and materializes only the raw data needed to answer that query. Over time, the structure adapts to the actual workload rather than paying full indexing cost before any exploration begins.