GT Baseball Strategy Visualization and Optimization

Two Step Stochastic Convex Linear Optimization

Inspiration

Our inspiration mainly came from the project description given to us by Georgia Tech Athletics. In the project description, there are diagrams from MLB Advanced Media, which provide a clear visual depiction of both pitches and bats.

What it does

Our deliverables visually present the path that a baseball travels after being pitched and after being hit by the batter. We created a software that takes data on batting data points and suggests optimal fielder placements.

How we built it

We built it using with Python using MatPlotLib and numpy and pandas to help process the given data. We used Gurobi and Pytorch. We did a lot of data exploration through monte carlo simulations and settled on approaching the problem as a mixed integer Linear Optimization Problem.

Challenges we ran into

The 3D graphing library within MatPlotLib gave us many challenges, as it was not as thoroughly supported as the 2D graphing library. There were many limitations, such as the inability to draw an arc in 3D. We had a really complex model and simplifying it to create lean code on gurobi was very difficult.

Accomplishments that we're proud of

We are proud of our visual depictions of the baseball, both after being pitched and after being batted. The GUI could use some work, but we are proud that it is both interactive for the user and connected to a backend which allows for easy parsing through the data.Our Simulation and Optimization program took a lot of time to think through and model. We arrived at the approach after trying many other data driven approaches first.

What we learned

We learned a lot about GUI in python and how to graphically depict paths in three dimension. We learned how to represent stochastic systems in programming.

What's next for GT Baseball Strategy Visualization and Optimization

We hope to land a contract with the GT athletics department to continue our visualization efforts and to present a final deliverable which is both easy to use and that provides useful insights to the GT baseball team. We hope to integrate concepts from graph theory and Markov decision processes to improve our strategy suggestion model. We would like to explore using reinforcement learning agents to make comparisons on natively discovered solutions to the mathematical ones.