Sidewater Basketball Analytics

SBA uncovers novel insights into basketball strategy through the use of high-resolution data. We first ask questions while channeling a deep, intuitive understanding of the mechanisms behind winning NBA basketball. Unsupervised Deep Learning Networks and simulation models are the foundation of our discovery process. Feature selection is the single greatest limitation of predictive algorithms. This is why we are lucky to live in 2018 - a time when models that use novel approaches like Reinforcement Learning are easily outperforming robust Neural Networks from 2017. The chief problem facing NBA data-crunchers is the Curse of Dimensionality. Regression models must transpose collected variables into higher-dimensional space to map an accurate predictive function. After one trains & fits their model, they are left with a predictive function in high-dimensional space. Dimensionality reduction works like a charm when you are telling a computer to look at something physical or auditory: "Computer, am I looking at a dog or a cat?" Our operating theory is that dimensionality reduction does not work on vectors, or eigenvectors, or any other numeric representation of tallied basketball events. In other words, no numeric representation other than points scored and points surrendered can flow through dimensional space without being numerically decontextualized.