Exploring Hierarchical Regression Models on NBA Shot Log Data
This was a group final project undertaken for Bayesian Machine Learning as part of the UVA MSDS program. Along with Edwin Purcell and Will McDevitt, I used NBA shot log data from the 2016-2017 season (Kaggle) to explore how well we could predict shot outcome (make versus miss) using Bayesian regression methods. We explored how well pooled, unpooled, and hierarchical models predicted shot outcome for both the Cleveland Cavaliers and Houston Rockets in the 2016-2017 season. For each model, we used both Variational Inference and Monte-Carlo Sampling to find the posterior distribution for each team and then compared the results across all models and teams. A paper summarizing our methods and results as well as a notebook walking through our code can be found below.