Kalaivani-g/NEW_RL_ENVIRONMENT

🎟️ Ticket Booking RL Environment (OpenEnv)

🚀 Overview

This project implements a Reinforcement Learning (RL) environment for a ticket booking system using the OpenEnv framework.

The environment simulates real-world challenges such as:

• —Limited seat allocation
• —Payment failures
• —Waitlist handling
• —Cancellation & reallocation
• —Priority-based booking

🎯 Objectives

• —Design a realistic RL environment
• —Ensure correct runtime behavior
• —Follow OpenEnv interface standards
• —Provide clear grading logic
• —Enable interpretable agent decisions

🧠 Features

✅ Core Booking Logic

• —Seat allocation under constraints
• —Waitlist system
• —Partial seat allocation
• —Cancellation & reallocation

💳 Payment Handling

• —Payment success / failure
• —Retry payment
• —Refund handling
• —Edge case: payment deducted but failed

🔥 Intelligence Layer

• —Priority-based allocation
• —Smart agent decision making
• —Handles conflicting requests

⭐ Explainability (Important)

Each step includes a reason field:

reason=Seats allocated successfully
reason=Payment retried successfully
reason=Already in waitlist

🏗️ Project Structure

ticket_env/
│
├── env.py
├── grader.py
├── inference.py
├── openenv.yaml
├── Dockerfile
├── requirements.txt
├── README.md
│
└── tasks/
    ├── easy.py
    ├── medium.py
    ├── hard.py

⚙️ Installation

pip install -r requirements.txt

▶️ Run the Environment

python inference.py

📊 Sample Output

[START] task=easy env=ticket_env model=baseline

[STEP] step=1 action={"action_type": "allocate", "user_id": "U1", "seats": 1} reward=0.45 done=false error=null reason=Seats allocated successfully

[STEP] step=2 action={"action_type": "allocate", "user_id": "U2", "seats": 1} reward=0.50 done=false error=null reason=Seats allocated successfully

[STEP] step=3 action={"action_type": "waitlist", "user_id": "U1", "seats": 0} reward=0.20 done=false error=null reason=Already in waitlist

[END] success=true steps=5 score=0.40 rewards=0.45,0.50,0.20,0.20,0.00

🧪 Tasks

Easy

• —All users have successful payments
• —Goal: efficient allocation

Medium

• —Includes failed payments
• —Requires retry + allocation strategy

Hard

• —Multiple users + limited seats
• —Requires prioritization + smart allocation

📈 Grading Logic

The environment evaluates performance using:

• —Allocation correctness (40%)
• —Payment handling (30%)
• —Efficiency (30%)

Score is normalized between 0 and 1.

🌐 Demo

👉 Hugging Face Space (UI Demo):

• —link coming soon *

💻 GitHub Repository

👉 GitHub link coming soon

🧠 Tech Stack

• —Python
• —OpenEnv
• —Gradio (for demo UI)

🏁 Conclusion

This project demonstrates:

• —A fully functional RL environment
• —Real-world scenario modeling
• —Clear evaluation metrics
• —Explainable agent behavior

🙌 Author

Kalaivani G Kamalee S Abirami M