Pitch Comp Tool

This month I am going in a bit of a different direction than my normal article in that we won’t be showing much analysis, just introducing a basic Shiny App I have made and how one could use it.

              A shiny app is an interactive web app powered by R that anyone can create. The whole concept of ours is to allow users to enter data on their release profile and fastball characteristics and then based on that we find MLB comps with good changeups, curveballs or sliders. The whole point of the app is to allow someone to see not just who they compare to, but who may be an ideal comparison to model a secondary pitch after based on their comps. This allows users to not just guess what secondary offering they want to design, but to have an educated idea of what works for them based on their arsenal.

              This app takes data from every MLB pitcher from 2015-2019 and on every pitch they threw. A user first enters their release height, side, extension, height in inches and handedness. Based on that we filter for players from our MLB data set that are a certain number of standard deviations from our input. We used manual standard deviations so we can control for each input how wide or narrow we define our scope for comps.

              The second part of our app takes a player’s fastball velocity and spin and the finds similar fastballs within that player’s release comps. This similarity is based on a scaled and centered Euclidean distance. Then, based on the top 50 fastball comps we return the top 20 players with the best SwStr% on one of their secondary offerings. This tells us what a player that is similar to our input may desire to develop.

              There are a few places where I am considering redefining how we find a “similar” player. At the most basic level I may want to add a pitch count filter, right now I have chosen to leave it off. For release profile we have a lot of similar metrics. For fastballs, we also have metrics (Velocity, Spin, Horizontal and Vertical Break) that all are related. Thus, simply using deviations or Euclidean distance here may be misleading. I may try PCA on the data and then do distance on the components. It also could help to find a way to weight the components most important to performance, however the goal here is to still say based on a player’s raw characteristics what may be possible.

              The last part of our app is a movement graph. This is nice because it can show you the type of pitch you may want to develop and how it may look shape wise. If you see all 20 or a large majority of your best comps based on your profile that have a sweeping slider that generates a lot of whiffs then this allows you to visualize that and get a pretty clear idea that a similar pitch could benefit you.

              To get started on the app all you have to do is enter what you want and also edit the filters if you would like. You can filter for a certain pitch type to comp on and a spin rate range for breaking balls if there is a certain space you think you may fall in. After entering everything, just hit the “Apply Filters” button and everything will calculate for you.

A link to the app can be found here