Lilli98/BeerGame-HeterogeneousVersion

🍺 Beer Game — Human & AI Collaboration

This is a Beer Game experiment interface built with Streamlit and hosted on Hugging Face Spaces. The goal is to study human behaviour in supply chain decision-making when collaborating with AI agents (Factory, Wholesaler, Retailer powered by OpenAI GPT-4o-mini).

🎮 Roles

• —Distributor: Human participant — you control the order quantity each week.
• —Factory / Wholesaler / Retailer: AI agents making decisions using LLMs.

📊 Features

• —Classic Beer Game settings:
• —36 weeks simulation
• —2-week shipping delay
• —1-week information delay
• —Standard customer demand sequence (constant → shock → fluctuations)
• —Human plays as Distributor, other roles are LLM agents.
• —Logs saved per participant and automatically pushed to Hugging Face Datasets Hub.
• —Simple Streamlit interface, deployable on Hugging Face Spaces.
• — ---

🛠️ Running the Experiment

1. On Hugging Face Spaces

1. 1.Deploy this repo as a Streamlit Space.
2. 2.Add Secrets (Settings → Secrets):
3. 3.OPENAI_API_KEY — your OpenAI API key
4. 4.Optional for S3 export:
5. 5.AWS_ACCESS_KEY_ID
6. 6.AWS_SECRET_ACCESS_KEY
7. 7.S3_BUCKET
8. 8.S3_REGION (default: us-east-1)

1. 1.Participants open the Space and either:
2. 2.Enter a custom participant ID, or
3. 3.Use the auto-generated ID shown in the app.

Example URL with participant ID:

https://huggingface.co/spaces/<user>/<space>?participant_id=ABC123

2. Locally

pip install -r requirements.txt
streamlit run app.py