running-experiments

What problem does it solve?

This Skill automates the execution and analysis of complex game theory experiments, allowing researchers and developers to efficiently test reinforcement learning agents and compare different algorithms.

Core Features & Use Cases

• Experiment Execution: Run predefined experiment types (Two-Player Symmetric, Different Cost, Different Ability, Three Players) with configurable parameters.
• Algorithm Comparison: Easily switch between PPO and gradient-based methods to compare their performance.
• Parameter Sweeping: Configure and run experiments across various noise parameters (q) and custom game settings.
• Analysis & Visualization: Facilitates the generation of convergence plots and detailed analysis of experiment results.
• Use Case: You want to test how a PPO agent learns the Nash equilibrium in a two-player tournament. You can use this Skill to launch the experiment, monitor its convergence, and generate plots comparing it to a gradient baseline.

Quick Start

Run a PPO experiment for the two-player symmetric case with q=40 and a specific seed.