rl-curriculum-bench 0.1.4

A benchmark suite for curriculum learning in multi-agent reinforcement learning

RL Curriculum Bench

A benchmark suite for curriculum learning in multi-agent reinforcement learning (MARL). This library provides standardized environments, complexity metrics, and curriculum strategies to facilitate reproducible curriculum learning research.

Features

• Environment Wrappers: Unified API for 3 PettingZoo environments (Pursuit, Cooperative Pong, Pistonball)
• Complexity Metrics: Graph-based and task-based metrics to measure environment difficulty
• Curriculum Strategies: Manual, performance-based, and complexity-based curriculum learning
• Evaluation Framework: Tools for comparing curriculum learning approaches
• GPU Support: Compatible with Stable-Baselines3 for efficient training

Installation

pip install rl-curriculum-bench

For RL training capabilities:

pip install rl-curriculum-bench[rl]

Quick Start

from rl_curriculum_bench.environments import make_env
from rl_curriculum_bench.metrics import compute_complexity
from rl_curriculum_bench.curriculum import ComplexityBasedCurriculum

# Create an environment
env = make_env('pursuit', difficulty='medium')

# Compute complexity metrics
metrics = compute_complexity('pursuit', difficulty='medium')
print(metrics)

# Create a curriculum
env_complexities = {
    'cooperative_pong': 1.5,
    'pursuit': 6.5,
    'pistonball': 7.5
}
curriculum = ComplexityBasedCurriculum(env_complexities, ascending=True)

# Get next environment in curriculum
next_env = curriculum.get_next_env()

Available Environments

• pursuit: Multi-agent pursuit-evasion task
• cooperative_pong: Two-player cooperative Pong
• pistonball: Cooperative ball manipulation with pistons

Each environment supports three difficulty levels: easy, medium, hard

Curriculum Strategies

Manual Curriculum

Fixed sequence of environments:

from rl_curriculum_bench.curriculum import ManualCurriculum

curriculum = ManualCurriculum(['env1', 'env2', 'env3'])

Performance-Based Curriculum

Progress when threshold is met:

from rl_curriculum_bench.curriculum import PerformanceBasedCurriculum

curriculum = PerformanceBasedCurriculum(
    ['env1', 'env2', 'env3'],
    threshold=0.75
)

Complexity-Based Curriculum

Automatically orders by complexity:

from rl_curriculum_bench.curriculum import ComplexityBasedCurriculum

curriculum = ComplexityBasedCurriculum(env_complexities, ascending=True)

Project Structure

rl_curriculum_bench/
├── environments/     # Environment wrappers
├── metrics/          # Complexity metrics
├── curriculum/       # Curriculum strategies
├── evaluation/       # Evaluation tools
└── utils/           # Utilities (config, seeds)

Requirements

• Python ≥3.8
• NumPy ≥1.20.0
• PettingZoo ≥1.24.0
• Gymnasium ≥0.26.0
• NetworkX ≥2.6.0
• PyYAML ≥5.4.0

License

MIT License

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.