Getting Started
Performance Benchmarks
Measured performance characteristics and memory profiles across dataset scales
Featuresmith is optimized for fast, predictable execution. The following performance profiles were measured directly on our standard Windows 11 AMD64 host running Python 3.13.7.
Hardware & Engine Specs
- OS Platform: Windows 11 (Architecture: AMD64)
- Runtime: Python 3.13.7 (or higher compatible)
- Vector Backends: Polars (vectorized lazy query planner) and pandas
Actual Performance Metrics
| Dataset Size (Rows) | Ingestion / Load (ms) | Profiling Engine (ms) | Rule Engine (ms) | End-to-End Audit (ms) | Peak Memory (MB) |
|---|---|---|---|---|---|
| 10,000 | 13.21 | 78.42 | 0.71 | 56.78 | 1.20 MB |
| 100,000 | 147.03 | 562.74 | 0.73 | 514.64 | 11.82 MB |
| 500,000 | 499.74 | 3,667.91 | 12.57 | 2,331.68 | 62.01 MB |
* Peak memory measures temporary heap allocations using tracemalloc. Rules execution operates instantly (under 10ms) because it audits statistical summaries already computed in memory.
Key Observations
- Linear Complexity: Execution times and memory footprint scale near-linearly with row count, taking under 4 seconds for half a million rows.
- Memory Gating: Peak memory remains highly constrained (only 62 MB for 500K rows) thanks to vectorized Polars engine allocations.