krna
Author & Maintainer
Sympodial Kawayan Rhizome Optimization Algorithm
A bamboo-inspired optimization algorithm published on PyPI. Lévy-flight exploration pairs with gradient exploitation, and the package ships with benchmarks, a multi-objective variant, and a scikit-learn tuner.
Overview
Most optimization algorithms live only in papers. krna is a working one. SKROA, the Sympodial Kawayan Rhizome Optimization Algorithm, is a swarm algorithm modeled on how running bamboo spreads, published as a real pip package with tests, docs, and reproducible benchmarks.
Architecture
Python 3.11+ with NumPy as the only core dependency. SKROA runs two phases. Agents explore with Lévy flights, then exploit with finite-difference gradients. Two extra operators keep the swarm healthy. Sympodial Clamping spreads out crowded agents, and Culm-Abortion prunes stalled ones and respawns them near the best point. MO-SKROA adds a non-dominated archive for multi-objective problems. The tuner accepts any scikit-learn model and searches mixed parameter spaces. Releases publish to PyPI through GitHub Actions with trusted publishing.
Outcomes
- Published on PyPI as krna, MIT licensed
- Benchmark suite runs Wilcoxon tests and reports honest results
- Every operator can be switched off for ablation studies
- Works as a hyperparameter tuner for scikit-learn models