project-proposal-2025

PokerLAB: A RESTful Environment for Reinforcement Learning in Poker

Abstract

Poker provides a classic and intriguing domain for RL research due to its incomplete information structure and strategic complexity. However, the lack of standardised hand history notations and parsing formats poses significant challenges, often leading researchers to expend valuable resources aligning data before experimentation. PokerLAB propose a robust and modular RESTful game engine designed specifically to streamline poker-based RL research.

Author

Name: Zhongtian Zheng

Student number: 48397494

Functionality

PokerLAB provides a flexible environment for RL in poker. It offers:

Scope

The minimum viable product will deliver:

Quality Attributes

1. Interoperability:

PokerLAB is on a journey to streamline RL training process for card games by prioritising interoperability. It aims to reduce the time researchers spend resolving data compatibility problems, making them to focus more fully on algorithm development. The system should seamlessly exchange data and integrate smoothly with existing datasets, models, and widely used third-party RL agents within the research community.

2. Modularity

Modularity ensures system remains manageable and maintainable. By clearly dividing the system into separate modules, PokerLAB an easily accommodate future extensions, modifications, and debugging, without impacting unrelated components.

3. Reliability

Scalability ensures that PokerLAB can support asynchronous and multi-threaded self-play training. It should be capable of handling simultaneous interactions from multiple agents while maintaining consistent performance, even under increased computational loads and concurrent requests.

Evaluation

1. Interoperability:

2. Modularity

3. Reliability