KuhnPoker
KuhnPoker
Description
KuhnPoker is an environment for evaluating agents on simplified poker with a 3-card deck. This environment wraps the KuhnPoker implementation from TextArena, a framework for text-based game environments.
Capabilities
- Strategic decision-making in imperfect information games
- Bluffing and probability-based reasoning
- Sequential decision-making with betting mechanics
- Two-player competitive game play against an LLM opponent
Compute Requirements
KuhnPoker does not require a sandbox. It has minimal compute requirements.
License
MIT.
Tasks
There are two splits: train (750 tasks) and test (750 tasks). Each split contains 50 tasks across each of 15 variants:
- KuhnPoker-v0
- KuhnPoker-v0-train
- KuhnPoker-v0-raw
- KuhnPoker-v0-short
- KuhnPoker-v0-short-train
- KuhnPoker-v0-short-raw
- KuhnPoker-v0-medium
- KuhnPoker-v0-medium-train
- KuhnPoker-v0-medium-raw
- KuhnPoker-v0-long
- KuhnPoker-v0-long-train
- KuhnPoker-v0-long-raw
- KuhnPoker-v0-extreme
- KuhnPoker-v0-extreme-train
- KuhnPoker-v0-extreme-raw
Each task is seeded for reproducibility.
Reward Structure
This is a sparse reward environment. Rewards are mapped from TextArena's native range of {-1, 0, 1} to {0.0, 0.5, 1.0} via (raw + 1) / 2.
We do not use LLM graders for this environment; reward is determined programmatically.
Data
Game state is generated procedurally by the TextArena engine using seeded randomness. No external data files are required.
Tools
Agents are given four tools:
check(): Check (pass without betting).bet(): Place a bet of 1 chip.call_bet(): Call the opponent's bet by matching it.fold(): Fold and surrender the pot.
Time Horizon
KuhnPoker is a multi-turn environment.
Environment Difficulty
Moderate. Kuhn Poker requires understanding poker fundamentals, probability estimation, and strategic betting decisions with imperfect information.
Other Environment Requirements
This environment requires an OpenAI API key (passed via secrets) to power the LLM opponent.
Safety
Agents in KuhnPoker interact only with a poker game simulation and have no access to external systems, the internet, or sensitive data. The environment does not present safety risks.
Citations
@software{textarena2024,
author = {Guertler, Leon and Banting, Wilfried and Pignatelli, Eduardo},
title = {TextArena},
year = {2024},
publisher = {GitHub},
url = {https://github.com/LeonGuertler/TextArena}
}