Welcome to Negotiation Platform’s documentation!

The Negotiation Platform is a comprehensive framework for conducting automated negotiations between Large Language Models (LLMs). This platform enables researchers to study negotiation behaviors, biases, and strategies in controlled environments.

User Guide:

Negotiation Platform:

Features

  • Multi-Game Support: Company car negotiations, resource allocation, integrative negotiations

  • LLM Integration: Support for various language models via Hugging Face

  • Bias Detection: Advanced statistical analysis tools for detecting negotiation biases

  • Configurable: Flexible YAML-based configuration system

  • Metrics: Comprehensive performance and fairness metrics

  • Analysis Tools: Built-in statistical analysis and visualization capabilities

Quick Start

  1. Install the package:

git clone https://github.com/tomz2000-lab/Masterthesis.git
cd Masterthesis
pip install -r requirements.txt

  1. Transfer project to a remote cluster (preserve directory structure)

# Using scp (PowerShell / Linux): replace <user> and <host> and <remote_path>
scp -r . <user>@<host>:/home/<user>/<remote_path>/Masterthesis

# Using rsync (recommended for large repos / resumable transfers):
rsync -avz --progress --exclude "negotiation_env/" --exclude ".git/" ./ <user>@<host>:/home/<user>/<remote_path>/Masterthesis

# Example (your earlier cluster):
scp -r . s123456@julia2.hpc.uni-wuerzburg.de:/home/s123456/Masterthesis

  1. Submit a batch run on the cluster using the provided Slurm scripts

# SSH into the cluster
ssh <user>@<host>

# Change to project directory
cd /home/<user>/<remote_path>/Masterthesis

# Inspect available slurm scripts (examples are in `batch_slurm_files/` and top-level)
ls -l batch_slurm_files/ run_*.slurm

# Submit a pre-provided batch job (adjust the script to set correct paths or env vars)
sbatch batch_slurm_files/run_company_car.slurm

# Or submit a generic single-game runner
sbatch run_single_game.slurm

# Use `squeue -u <user>` to check job status, and `sacct -j <jobid>` to inspect job output

  1. Transfer result files from cluster to local machine

# Wait for jobs to complete, then transfer result files back to local machine
# Download the generated .out files from the cluster
scp <user>@<host>:/home/<user>/<remote_path>/Masterthesis/batch_comparison/ ./batch_comparison/ -r

# Or download specific result files
scp <user>@<host>:/home/<user>/<remote_path>/Masterthesis/batch_comparison/resource_game/resource_allocation_*.out ./batch_comparison/resource_game/

# Example with specific job output:
scp s123456@julia2.hpc.uni-wuerzburg.de:/home/s123456/Masterthesis/batch_comparison/resource_game/resource_allocation_2037717.out ./batch_comparison/resource_game/

  1. Run win and metric comparisons locally

# Run win statistics analysis on specific result file
python results/win_statistics.py batch_comparison/resource_game/resource_allocation_2037717.out

# Run metrics statistics analysis on the same file
python results/metrics_statistics.py batch_comparison/resource_game/resource_allocation_2037717.out

# For company car results:
python results/win_statistics.py batch_comparison/company_car/company_car_1862304.out
python results/metrics_statistics.py batch_comparison/company_car/company_car_1862304.out

# For integrative negotiation results:
python results/win_statistics.py batch_comparison/integrative_game/integrative_negotiation_1858216.out
python results/metrics_statistics.py batch_comparison/integrative_game/integrative_negotiation_1858216.out

Notes

  • Replace placeholder usernames, hostnames and paths with your cluster credentials and target directories.

  • Ensure the cluster has required Python packages installed or create a virtual environment and install requirements.txt there.

  • If using GPUs, load the appropriate modules (CUDA, etc.) before running sbatch or inside your Slurm scripts.

  • Slurm scripts in batch_slurm_files/ may need path adjustments for the Python executable or the virtual environment activation commands.

Indices and tables