Game-Agnostic Value Functions through Automatic JSON Feature Extraction

2026-08-30

AI Summary

"This study addresses the generalizability of value function training in game environments, proposing a novel table game evaluation framework called JSON-Bag VF. This approach transforms the game process into a sequence of tokens in JSON format, which are then used to train a game-agnostic value function. Additionally, the study incorporates a random forest-based feature selection technique to optimize model performance. Experimental results across six different table games demonstrate that JSON-Bag OSLA outperforms traditional OSLA agents, underscoring the importance of appropriate feature selection in enhancing model effectiveness."

Abstract

JSON Bag-of-Tokens (JSON-Bag) is a recently proposed method to generically represent game trajectories by tokenizing their JSON descriptions. We introduce JSON-Bag VF, a game-agnostic approach to training value functions for game-playing agents using JSON-Bag prototypes. We show that this approach can be enhanced with Random Forest-based feature selection and a method to select game-stage-specific features. We evaluate JSON-Bag VF with One-step-look-ahead (JSON-Bag OSLA) on six tabletop games over different combinations of prototype-tokenization and feature selections. JSON-Bag OSLA outperforms baseline OSLA agents in most games. Our analysis also shows that feature selection significantly improves JSON-Bag VF and that feature selection is the most important factor in JSON-Bag VF performance, over prototype-tokenization.

Problem

• Game-agnostic
• Value Functions
• JSON Feature Extraction
• Feature Selection

Innovation

• JSON-Bag VF
• Random Forest-based feature selection
• game-stage-specific features