AMBER (Agent-based Modeling with Blazingly Efficient Records)

AMBER is a Python framework for agent-based modeling that uses Polars for efficient data handling and analysis. AMBER provides a clean, robust API for creating parallel, high-performance simulations in Python.

🚀 Performance

AMBER achieves high performance through its columnar memory layout (SoA), KD-Trees, and SIMD-vectorized operations.

Benchmark vs Other Frameworks (5,000 Agents, 100 Steps)

Framework	Language	Architecture	Speed Rank
Agents.jl	Julia	Vectorized	🥇 1st (0.02s - 0.9s)
AMBER	Python	Vectorized	🥈 2nd (0.1s - 2.3s)
SimPy (Dense)	Python	Process/DES	🥉 3rd (0.3s - 4.3s)
Melodie	Python	Hybrid	4th (0.4s - 20s)
AgentPy	Python	Object	5th (2s - 30s)
Mesa	Python	Object	6th (50s - 30s)

*SimPy is exceptionally fast for sparse models (like SIR) due to its event-driven nature, but AMBER is faster for dense/movement-heavy models.

AMBER is the state-of-the-art for dense simulation in Python, while SimPy offers an alternative for event-driven logic.

🚀 Quick Start

import ambr as am

# Define a custom agent
class MyAgent(am.Agent):
    def setup(self):
        self.value = self.p.initial_value
        
    def step(self):
        self.value += self.model.random.randint(-1, 2)
        self.record('value', self.value)

# Define a model
class MyModel(am.Model):
    def setup(self):
        self.agents = am.AgentList(self, self.p.agents, MyAgent)
        
    def step(self):
        self.agents.call('step')
        
    def update(self):
        super().update()
        self.agents.record('value')

# Run a simulation
parameters = {
    'agents': 10,
    'initial_value': 5,
    'steps': 100
}

model = MyModel(parameters)
results = model.run()

⚡ Advanced: Vectorized Updates

For maximum performance, access the underlying Population manager to perform SIMD-vectorized state updates, bypassing Python loop overhead:

def step(self):
    # Create a batch update context
    with self.population.create_batch_context() as batch:
        # Queue updates logic (executed in Rust/Polars)
        batch.add_update(target_ids, 'wealth', 1)
        batch.add_update(source_ids, 'wealth', -1)
    
    # State is applied atomically here

🔬 Optimization

AMBER includes powerful optimization capabilities for parameter tuning:

from ambr.optimization import ParameterSpace, grid_search

# Define parameter space
parameter_space = ParameterSpace({
    'agents': [10, 50, 100],
    'initial_value': [1, 5, 10],
    'steps': 100
})

# Run optimization
results = grid_search(MyModel, parameter_space, 'some_metric')
best_params = results[0]['parameters']

📦 Installation

pip install ambr

🏗️ Features

• Simple API: Intuitive interface for agent-based modeling
• High Performance: Efficient data handling with Polars DataFrames
• Optimization: Built-in parameter optimization with grid search, random search, and Bayesian optimization
• Environments: Support for grid, network, and continuous space environments
• Experiments: Run multiple simulations with parameter sampling
• Random Number Generation: Reproducible simulations with controlled randomness

📚 Examples

Working examples are available in the examples/ directory:

• Wealth Transfer Model: Economic inequality simulation
• Virus Spread Model: Epidemiological SIR model
• Flocking Simulation: Boids flocking behavior
• Forest Fire Model: Cellular automata fire spread
• Network Simulations: Graph-based agent interactions

📖 Documentation

• Documentation: https://ambr.readthedocs.io/
• Paper: https://arxiv.org/abs/2601.16292

📝 How to cite?

If you use AMBER in your research, please cite our paper:

@article{pham2026amber,
  title={AMBER: A Columnar Architecture for High-Performance Agent-Based Modeling in Python},
  author={Pham, Anh-Duy},
  journal={arXiv preprint arXiv:2601.16292},
  year={2026}
}

🤝 Contributing

We welcome contributions! Please see our contributing guidelines for more information.

📄 License

This project is licensed under the BSD 3-Clause License - see the LICENSE file for details.