negmas-genius-agents

Python reimplementations of 141 Genius negotiating agents from ANAC competitions 2010-2019 for use with the NegMAS framework.

IMPORTANT NOTICE: AI-ASSISTED IMPLEMENTATION

The agents in this package were reimplemented from Java to Python with the assistance of AI (Large Language Models). While efforts have been made to faithfully reproduce the original agent behaviors, these implementations may not behave identically to the original Genius agents in all cases.

If you require guaranteed behavioral equivalence with the original Java implementations, please use the GeniusNegotiator wrapper in NegMAS, which runs the actual Java agents via a bridge.

Bug reports and contributions to improve behavioral fidelity are welcome.

Genius Attribution

This package provides Python reimplementations of agents originally developed for Genius (General Environment for Negotiation with Intelligent multi-purpose Usage Simulation) — a Java-based automated negotiation framework developed at TU Delft.

• Original Source: http://ii.tudelft.nl/genius/
• Original License: GPL-3.0
• Original Version: 10.4

Genius has been the standard platform for the ANAC (Automated Negotiating Agents Competition) since 2010. The agents in this package have been reimplemented in Python to provide native NegMAS compatibility without requiring Java.

If you use this package, please cite Genius:

@article{lin2014genius,
  title={Genius: An integrated environment for supporting the design of generic automated negotiators},
  author={Lin, Raz and Kraus, Sarit and Baarslag, Tim and Tykhonov, Dmytro and Hindriks, Koen and Jonker, Catholijn M},
  journal={Computational Intelligence},
  volume={30},
  number={1},
  pages={48--70},
  year={2014},
  publisher={Wiley Online Library}
}

Features

• 141 ANAC agents from competitions 2010-2019
• 5 basic time-dependent agents (Boulware, Conceder, Linear, Hardliner)
• Pure Python - No Java dependency required
• Seamless integration with NegMAS mechanisms and tournaments
• Full compatibility with NegMAS utility functions and outcome spaces
• Easy agent retrieval with get_agents() function

Installation

pip install negmas-genius-agents

Or with uv:

uv add negmas-genius-agents

Requirements

• Python >= 3.10
• NegMAS >= 0.10.0

Development Installation

git clone https://github.com/autoneg/negmas-genius-agents.git
cd negmas-genius-agents
uv sync --dev

Quick Start

from negmas import SAOMechanism, make_issue
from negmas.preferences import LinearAdditiveUtilityFunction as U

from negmas_genius_agents import Atlas3, AgentK, get_agents

# Create negotiation scenario
issues = [make_issue("price", (0, 100)), make_issue("quality", (0, 10))]
mechanism = SAOMechanism(issues=issues, n_steps=100)

# Create agents with utility functions
ufun1 = U.random(issues=issues)
ufun2 = U.random(issues=issues)

# Use ANAC competition winners
agent1 = Atlas3(ufun=ufun1)   # ANAC 2015 winner
agent2 = AgentK(ufun=ufun2)   # ANAC 2010 winner

# Run negotiation
mechanism.add(agent1)
mechanism.add(agent2)
result = mechanism.run()

print(f"Agreement: {result.agreement}")
print(f"Agent1 utility: {ufun1(result.agreement)}")
print(f"Agent2 utility: {ufun2(result.agreement)}")

Available Agents

Summary

Using get_agents()

from negmas_genius_agents import get_agents

# Get all agents (129 total)
all_agents = get_agents()

# Get all ANAC winners
winners = get_agents(category="winners")

# Get agents from a specific year
agents_2015 = get_agents(group="anac2015")

# Get finalists (top 3) from multiple years
early_finalists = get_agents(group=["anac2010", "anac2011"], category="finalists")

# Get basic time-dependent agents
basic = get_agents(group="basic")

Basic Time-Dependent Agents

ANAC 2010 Agents

ANAC 2011 Agents

ANAC 2012 Agents

ANAC 2013 Agents

ANAC 2014 Agents

ANAC 2015 Agents

ANAC 2016 Agents

ANAC 2017 Agents

ANAC 2018 Agents

ANAC 2019 Agents

Mixing with NegMAS Agents

Genius agents can negotiate with native NegMAS agents:

from negmas.sao import AspirationNegotiator
from negmas_genius_agents import HardHeaded

mechanism = SAOMechanism(issues=issues, n_steps=100)
mechanism.add(HardHeaded(name="genius_agent"), preferences=ufun1)
mechanism.add(AspirationNegotiator(name="negmas_agent"), preferences=ufun2)

state = mechanism.run()

Running Tournaments

from negmas.sao import SAOMechanism
from negmas_genius_agents import get_agents

# Get all ANAC winners
winners = get_agents(category="winners")

# Run round-robin tournament
results = []
for i, AgentA in enumerate(winners):
    for AgentB in winners[i+1:]:
        mechanism = SAOMechanism(issues=issues, n_steps=100)
        mechanism.add(AgentA(name="A"), preferences=ufun1)
        mechanism.add(AgentB(name="B"), preferences=ufun2)
        state = mechanism.run()
        results.append({
            "agent_a": AgentA.__name__,
            "agent_b": AgentB.__name__,
            "agreement": state.agreement is not None,
        })

Development

Running Tests

uv run pytest tests/ -v

Building Documentation

uv run mkdocs serve

License

AGPL-3.0 License - see LICENSE for details.

Acknowledgments

• NegMAS - Negotiation Managed by Situated Agents
• Genius - General Environment for Negotiation with Intelligent multi-purpose Usage Simulation
• ANAC - Automated Negotiating Agents Competition

Citation

If you use this library in your research, please cite this package, NegMAS, and Genius:

@software{negmas_genius_agents,
  title = {negmas-genius-agents: Python Reimplementations of Genius Negotiating Agents},
  author = {Mohammad, Yasser},
  year = {2024},
  url = {https://github.com/autoneg/negmas-genius-agents}
}

@inproceedings{mohammad2022negmas,
  title={NegMAS: A Platform for Automated Negotiations},
  author={Mohammad, Yasser and Nakadai, Shinji and Greenwald, Amy},
  booktitle={Proceedings of the 21st International Conference on Autonomous Agents and Multiagent Systems},
  pages={1845--1847},
  year={2022}
}

@article{lin2014genius,
  title={Genius: An integrated environment for supporting the design of generic automated negotiators},
  author={Lin, Raz and Kraus, Sarit and Baarslag, Tim and Tykhonov, Dmytro and Hindriks, Koen and Jonker, Catholijn M},
  journal={Computational Intelligence},
  volume={30},
  number={1},
  pages={48--70},
  year={2014},
  publisher={Wiley Online Library}
}

Related Projects

• negmas-negolog - NegMAS wrappers for NegoLog agents
• negmas - The NegMAS negotiation framework