Lastly, I put together a Shiny app for the 2024 season so you can see the predicted swing types for all 650 hitters, along with a summary of how each of their swings work and what the results are

I also look at predicting hitter specific swing types given location + count, and see how hitters differ in their approach

New article today about work I’ve done on classifying swing types and how hitters vary their swing based on location and count. Personally, I’ve found it hard to group swings together in my head based on the new data, so this attempts to do that.

Read this nice article on Shape Constrained Additive Models yesterday. Tried it out today on some generated data, comparing a standard GAM to a SCAM with upward trend (correct) and SCAM with downward trend (incorrect). GAM and correct SCAM are similar, with the SCAM being less wiggly.

An issue i have found is that this method is extremely sensitive to the hazard function. Any experts know how to make this less jumpy? Below is Arjun Nimmala's k rate using expected changepoint of 10 games and 50, and as you can see there is an extreme difference

Last winter break project, I put together a Shiny App to guess the Stuff+ for a given pitch. Hoping this helps people gain intuition on how Stuff models work while having some fun. Take a look! david-gerth.shinyapps.io/stuff_plus_g...

Experimenting a bit with using changepoint analysis on in-game data

fwiw here's the list of 17 players with confidence intervals not containing 0.

A bit surprised by this result, but I wanted to see how much the "Big 3" of hitting (Bat Speed, Bat to Ball, Swing Decisions) explained pull fly ball rate. Turns out, basically nothing. Looking just at 2024.

Here's my HOF ballot