MCTS Pipeline Search
Optimizes machine learning pipelines and hyperparameter combinations using Monte Carlo Tree Search with UCB1 exploration.
MCTS Pipeline Search guides Claude Code in systematically exploring machine learning pipelines, neural architectures, ensemble weights, and hyperparameter configurations using Monte Carlo Tree Search. By leveraging the UCB1 algorithm to balance exploitation of top-performing configurations with exploration of unvisited branches, it intelligently prunes dead ends where grid search or random search would exhaust computational budgets. This skill is ideal for competitive Kaggle pushes, multi-dimensional search spaces, and automated ML pipeline discovery.
主要功能
01Enables custom parameter expansion and continuous neighborhood tuning
02Balances hyperparameter exploration and exploitation using the UCB1 algorithm
03Prunes unpromising pipeline branches to maximize compute budget efficiency
04Provides best-practice guidance on tree depth, iteration counts, and caching
055 GitHub stars
06Supports custom search DAGs across models, feature sets, and ensemble architectures
使用场景
01High-Dimensional Hyperparameter Tuning: Exploring large search spaces with 4+ dimensions more efficiently than standard grid search.
02Competitive ML / Kaggle Top-1% Push: Systematically searching multi-model ensemble architectures and feature combinations beyond brute force limits.
03Automated ML Pipeline Design: Discovering non-obvious combinations of feature preprocessing, model algorithms, and post-processing steps.