ix-bandit

name	ix-bandit
description	Multi-armed bandit simulation — epsilon-greedy, UCB1, Thompson sampling
disable-model-invocation	true

Multi-Armed Bandits

Simulate bandit algorithms to find the best arm given uncertain rewards.

When to Use

When the user needs to compare exploration-exploitation strategies, simulate A/B testing, or understand bandit algorithms.

Capabilities

• Epsilon-greedy — Explore with probability ε, exploit otherwise
• UCB1 — Upper Confidence Bound (optimism in the face of uncertainty)
• Thompson sampling — Bayesian approach with posterior sampling

Key Concepts

• Higher true mean → algorithm should pull that arm more
• UCB1 gives logarithmic regret bounds
• Thompson sampling is often best in practice

Programmatic Usage

use ix_rl::bandit::{EpsilonGreedy, UCB1, ThompsonSampling};

MCP Tool

Tool name: ix_bandit Parameters: algorithm, true_means, rounds, epsilon