This chapter explores the potential of learning-based agent models in play evaluation for sports analytics. It begins by discussing the fundamentals of agent modeling, including key concepts in this field, such as the Markov decision process, reinforcement learning, and game theory, and including several simulation platforms. Next, the inverse approach for player and team evaluation from data and the forward approach for virtual simulations are introduced, and the advantages of using learning-based methods over traditional simulation techniques are highlighted. These approaches collectively contribute to the advancement of modeling complex scenarios, evaluating different tactical choices, and suggesting optimal actions in sports. Finally, this chapter examines the technical and practical challenges associated with these methods, as well as future research opportunities in the field. This chapter may provide us with a comprehensive understanding of how learning-based agent modeling can enhance play evaluation and contribute to the advancement of sports analytics.