Data Scientist at Intuit/Mailchimp. I like to share random musings on R, Stats, and College Football
Atlanta, GA
Joined June 2010
- Tweets1.9K
- Following386
- Followers410
- Likes2.5K
Pinned Tweet
I've uploaded some CFB Data for open source use github.com/mattmills49/CFB_A… 10 years of
Team recruit rankings
Draft Picks
Schedule and Results
You can also adapt any gbm to predict multiple, parametrized outputs, not just multiple classification predictions statmills.com/2026-04-06-gra…
Checkout the latest XGBoost update "Introducing the XGBoost Vector-Leaf Model" from jiamin yuan and Rory Mitchell
xgboost.ai/2026/08/25/introd…
Very cool to find a well written statistics blog post like we used to see all the time on this app: @bruno_nicenboim giving a walkthrough of Ordinal Models with {brms} bruno.nicenboim.me/posts/pos… #rstats
CFB/NFL Legend Ndamukong Suh going full YIMBY belongs in @TheZvi ‘s next housing roundup.
You can extend Gradient Boosting to fit many more models than just target predictions. My blog post from earlier this week walks through how you can fit the coefficients of smoothing splines with Gradient Boosting statmills.com/2026-04-06-gra…
I have a new blog post out today that I'm really excited about. I walk through how you can use Gradient Boosting to fit entire vectors of parameters for each observation, not just a single prediction.
The result is smooth curves that can learn high dimensional interaction effects that you can fit at scale!
I hope people find this walkthrough useful! statmills.com/2026-04-06-gra…
I've got a new blog post out about how to do proper Data Science in the age of LLMs. My thesis is that DS is a multiplicative process which separates it from more traditional software dev; if one assumption is off then the result is wrong in a way it isn't with a UI (1/3)
My biggest struggle is that AI can produce code that runs but may violate an assumption about the data; observations get dropped, duplicated, or mis-aligned without you knowing (2/3)
This is a very *rough* framework I know, but it was helpful for me to think about it in this way to figure out what tools and processes I needed to build to improve the generated code I use for data analysis. I hope its helpful to others as well. statmills.com/2025-05-03-dat…
10 of the 11 writers list FSU as making the playoff, none of them include Clemson at all, and yet at Fanduel right now Clemson is still the favorite to win the ACC at +185 sportsbook.fanduel.com/navig…
College Football Playoff predictions: Who's most likely to make the field - via @ESPN App espn.com/college-football/st…
Fun wrinkle for GT this year; only 3 conference opponents are playing better than expected. Technically FPI has us favored in every game until UGA lol @FTRSJoey
I'd guess you'd find the same results in basketball as well, the game the same it just got more fierce. youtube.com/watch?v=fp4but75…
NEW with @KuperSimon
The prevailing narrative around increased injuries and player workload in elite football is wrong.
Players don’t play more football than in the past. What has changed is a sharp rise in intensity of play.
Not more minutes, but each minute exerts more load.
Sharing for the morning crowd; My latest blog post covers how you can fit shape constrained models in python leveraging splines and JAX
My latest blog post is a walk-through of how Shape Constrained P-splines work and how you can use them to fit a curve of any arbitrary shape like monotonically increasing or decreasing
#pydata #pystats #datascience #MachineLearning
My latest blog post is a walk-through of how Shape Constrained P-splines work and how you can use them to fit a curve of any arbitrary shape like monotonically increasing or decreasing
#pydata #pystats #datascience #MachineLearning
This means you can enforce arbitrary shapes, even convex and concave, but still leverage all the benefits of a traditional GAM. Even better they are so straightforward you can fit them using general optimization packages like {jax} and {scipy}
Hopefully some of y'all find it useful! statmills.com/2025-05-03-mon…