Back to feed
arXiv cs.AI·

JSON-Bag: A generic game trajectory representation

Signal
72
Hype
15
In three linesJSON-Bag tokenizes JSON descriptions of game trajectories and uses Jensen-Shannon distance for comparison. Tested on 6 tabletop games (7 Wonders, Dominion, Connect4, etc.), the model outperforms baselines on agent, parameter, and seed classification. Efficient in few-shot settings and enables automatic feature extraction.
Read source
Your take?
BenchmarksPapers

Summary generated by Claude — human-verified