Perceptual Similarity for Measuring Decision-Making Style and Policy Diversity in Games

TL;DR

Three enhancements to increase accuracy are introduced to measure playstyle similarity based on game screens and raw actions: multiscale analysis with varied state granularity, a perceptual kernel rooted in psychology, and the utilization of the intersection-over-union method for efficient evaluation.

Abstract

Defining and measuring decision-making styles, also known as playstyles, is crucial in gaming, where these styles reflect a broad spectrum of individuality and diversity. However, finding a universally applicable measure for these styles poses a challenge. Building on Playstyle Distance, the first unsupervised metric to measure playstyle similarity based on game screens and raw actions, we introduce three enhancements to increase accuracy: multiscale analysis with varied state granularity, a perceptual kernel rooted in psychology, and the utilization of the intersection-over-union method for efficient evaluation. These innovations not only advance measurement precision but also offer insights into human cognition of similarity. Across two racing games and seven Atari games, our techniques significantly improve the precision of zero-shot playstyle classification, achieving an accuracy exceeding 90 percent with fewer than 512 observation-action pairs, which is less than half an episode of these games. Furthermore, our experiments with 2048 and Go demonstrate the potential of discrete playstyle measures in puzzle and board games. We also develop an algorithm for assessing decision-making diversity using these measures. Our findings improve the measurement of end-to-end game analysis and the evolution of artificial intelligence for diverse playstyles.

Figures (18)

• Figure 1: Illustration of the Playstyle Distance computation using a hierarchical discrete state encoder $\phi$. The Venn diagram highlights the intersection of discrete states for distance calculation.
• Figure 2: (a) Degree of Similarity: This demonstrates how multiple candidate points C can share identical distance values from a target point T, emphasizing that as distance increases, the degree of similarity information diminishes. (b) From Playstyle Distance to Playstyle Similarity: This transformation begins by processing an observation sample into multiple discrete states of varying granularity. A perceptual kernel then transforms these distance values into probabilistic similarities, using the concept of overlapping regions for intuitive understanding. Lastly, the application of the intersection-over-union method refines the similarity based on the Jaccard index, enhancing measurement comprehensiveness across all observed data.
• Figure 3: Three game platforms and the illustration of zero-shot playstyle classification tasks.
• Figure 4: Comparison of Efficacy: Probabilistic vs. Distance Approaches. The plot illustrates the relationship between accuracy (Y-axis) and size of the sampled observation-action pairs (X-axis). The shaded area indicates the range between min and max accuracy among three encoder models.
• Figure 5: Playstyle Measure Evaluation in TORCS, RGSK, and Atari Console. The plots showcase the efficacy of different measures in the context of the "Full Data Evaluation" subsection. The shaded area indicates the range between min and max accuracy among three encoder models.