Rust-Based Random Forest 'Fru' Outperforms scikit-learn by Hundreds of Times
kpiwonski · reddit · 2026-08-11
Researchers have published a highly optimized Rust-based implementation of Random Forest called Fru in the Software X journal, complete with bindings for both Python and R.
In Python environments, Fru outperforms the widely used scikit-learn implementation by several factors, reaching hundreds of times faster in specific scenarios. For R, it is typically a few dozen percent faster than the ranger package, with speedups reaching several times depending on the use case.
Additionally, the tool features a novel implementation of permutation importance for an extra performance boost. Thanks to its layered design, the Python version utilizes Arrow PyCapsule to integrate seamlessly with popular data libraries like pandas, polars, and pyarrow.
More from Research
- Tiny-task benchmarks no longer make sense for frontier models, argues dev — pvncher · 2026-09-22
- Why collaborating agents beat subagents: broadcasting breakthroughs in parallel — IgorCarron · 2026-09-22
- Frank Noe Clarifies How Their AI Solves the Electronic Schrödinger Equation — FrankNoeBerlin · 2026-09-22
- Did OpenAI Solve the Wrong Navier-Stokes Problem? Scientific American Weighs In — scientificamerican · 2026-09-22
- Harvard open sources LLM inference traces: 6.12 billion requests from a year of production traffic — markjeffrey · 2026-09-22
- Zhejiang U & SJTU unveil DAS, an agent that writes publication-ready surveys in an hour — jiqizhixin · 2026-09-22