matrix-profile-rs

Phase 2/5 (50%)

← Back to Projects

The Insight

Developing…

Quick Facts

What This Is

A high-performance Rust implementation of Matrix Profile algorithms for time series analysis. Matrix Profiles enable pattern discovery, anomaly detection, and similarity search in univariate time series without domain knowledge or parameter tuning.

Think of it as “find repeating patterns and anomalies in any time series data” with a simple API: df.select(pl.col("ts").mp.stomp(m=20)) for Polars users, or direct Rust APIs for maximum performance.

Problem It Solves

Time series analysis requires identifying:

• Repeating patterns (motifs): “This sensor pattern happened 15 times before failure”
• Anomalies (discords): “This heartbeat segment is unlike any other”
• Similar segments: “Find all sequences similar to this known good pattern”

Existing solutions:

• Python libraries (stumpy): Slow, JIT dependencies, awkward array manipulation
• Manual implementation: Complex algorithms, easy to get wrong, poor performance
• Academic prototypes: Not production-ready, missing ergonomics

matrix-profile-rs provides production-quality implementations with:

• Native performance (no JIT warmup)
• Clean APIs (.motifs(k=3) instead of array indexing)
• Easy distribution (single binary, no Python runtime)
• Polars integration (treat Matrix Profiles as DataFrame operations)

Features

Phase 1: Parallel Core Engine (25% complete)

• ✅ STOMP implementation with QT recurrence
• ✅ Basic parallel execution via Rayon
• 🔄 SCAMP kernels for numerical stability
• 🔄 Exclusion zone logic

Phase 2: Discovery Ergonomics (In Progress)

• 🔄 MatrixProfile result type with metadata
• ⏳ .motifs(k) and .discords(k) methods
• ⏳ Interpretable output structures

Phase 3: Anytime Insights (Planned)

• ⏳ SCRIMP++ for progressive refinement
• ⏳ Early stopping with confidence bounds

Phase 4: Hardware Optimization (Planned)

• ⏳ SIMD vectorization (AVX2/NEON)
• ⏳ Memory-efficient tiling for large datasets

Phase 5: Ecosystem Integration (Planned)

• ⏳ Polars plugin with .mp.stomp() syntax
• ⏳ Zero-copy result conversion
• ⏳ Python bindings via PyO3

Architecture

Algorithm Stack

• STOMP: Iterative exact algorithm using QT recurrence for sliding dot products
• SCAMP: Parallel exact algorithm optimized for multi-core CPUs
• SCRIMP++ (planned): Anytime algorithm providing approximate results that improve over time

Data Flow

Time Series Data
    ↓
Matrix Profile Calculation (STOMP/SCAMP/SCRIMP++)
    ↓
MatrixProfile struct (distances, indices, metadata)
    ↓
Discovery APIs → motifs() / discords() / to_polars()

Key Components

• Core Engine: Parallel computation using Rayon, SIMD-optimized kernels
• Result Type: Binding-friendly MatrixProfile with Vec-backed buffers
• Discovery API: High-level methods for pattern extraction
• Polars Plugin (Phase 5): Native integration with Polars DataFrames

Use Cases

Predictive Maintenance:

• Find recurring degradation patterns before failure
• Detect anomalous sensor behavior

Healthcare:

• Identify irregular heartbeat patterns
• Find repeating movement patterns in activity data

Finance:

• Discover recurring market microstructures
• Detect anomalous trading patterns

Operations:

• Find repeating load patterns for capacity planning
• Detect anomalous system behavior