google-deepmind/open_spiel
Open Spiel
Open Spiel is a research library and framework for reinforcement learning, planning, and multi-agent game simulation. It provides a system for representing single-agent and multi-agent games across zero-sum, cooperative, and imperfect information scenarios.
The project utilizes a standardized abstract game interface to decouple game logic from algorithms, allowing agents to run across different game types. It implements performance-critical logic in C++ with Python bindings and uses deterministic seeded simulation to ensure reproducibility for research benchmarking.
The framework covers a broad range of capabilities including computational search and planning algorithms, multi-agent reinforcement learning, and game theory research. It also includes tools for learning dynamics analysis to track agent behavior and calculate evaluation metrics.
Features
- Game Simulation Environments - Implements a system for simulating single-agent and multi-agent games for reinforcement learning and behavioral research.
- Game Search and Planning - Applies computational search and planning techniques to identify the best sequence of moves in complex game states.
- Game Theory Programming - Implements computational models for zero-sum, cooperative, and imperfect information games to study strategic decision making.
- Reinforcement Learning Simulators - Provides a framework for running and reproducing reinforcement learning and search algorithms in controlled game environments.
- Reinforcement Learning Research Frameworks - Provides a comprehensive framework for prototyping and evaluating reinforcement learning and search algorithms across game environments.
- Multi-Agent Training - Provides tools for training and evaluating agents that interact in shared environments through competition or cooperation.
- Game Logic Decouplers - Ships a standardized API layer that separates agent logic from game mechanics to support diverse game types.
- Game State Feature Encoding - Encodes game history and board states into numerical tensors for use in neural networks.
- Machine Learning Evaluation - Ships tools for calculating evaluation metrics and tracking agent behavior to identify performance trends during learning.
- Reinforcement Learning Algorithm Analyzers - Includes tools for comparing the performance and convergence of different learning algorithms using standardized metrics.
- Research Libraries - Provides a research library of planning algorithms and evaluation metrics to analyze agent learning dynamics.
- Algorithm Implementations - Provides a library of decoupled search and learning algorithm implementations that interact with game environments.
- Python-C++ Bindings - Utilizes C++ wrappers to expose performance-critical game logic to Python for efficient simulation.
