
With the help of LLMs, I have partially solved a long-troublesome issue with GBDTs! 1. Early stopping is a nice, clean way to automatically tune your GBDT 2. You by default leave some performance on the table with your hold-out set ...

currently: data science / optical @ NFL prev: data science @ GSW sometimes bball analytics moreso on bsky: bsky.app/profile/bbstats.bsk…

With the help of LLMs, I have partially solved a long-troublesome issue with GBDTs! 1. Early stopping is a nice, clean way to automatically tune your GBDT 2. You by default leave some performance on the table with your hold-out set ...

In summary: Models can be good at predicting and have uninterpretable coefficients
No, but I've played around with it (a joint-fitting DPM model), all fit in one stage. This is similar to what Goerge Berry does. You get something akin to wins produced, with bigs meaningfully overrated. It predicts reasonably well, but doesn't generalize super well to new lineups in my testing, and less interpretable/credible ratings.

What's changed in this year's NBA? Examining early-season trends like shot location, pace, net rating, and more to previous seasons. Plus, as you can see in the thumbnail: I got to use a box-and-whisker plot! Huzzah! basketballpoetry.com/p/whats…

indeed - my main shock was that it came on the Kings! the steal rate will almost certainly drop but - his deflections are good too.
#25 in Defensive Net Pts/100 Poss. Defends his matchup well but also defending well as a help guy, which can be difficult to do. Especially on the third worst defensive team in the league.

OpenRAPM believes in Ty (as did I in a past life, if you check my resume)

It's early but - Leaky Black is #22 in OpenRAPM's defensive per-100 rating this season Per-36 defensive numbers: 2.9 STL, 1.0 BLK, 4.7 DRB openrapm.com #STEALZZZ

big walk is a) the best game I have ever played and b) the most 'analog' feeling game I've ever played it is the least AI-slop thing imaginable
Big Walk is being review bombed after House House revealed AI was used on “well under 0.1%” of its code The studio says a chatbot helped fix a Unity bug, with the final solution using around 10 lines of heavily modified code


I also adjust for the small but significant amount that you get played on offense vs defense! (Shoutout to GP2)

OpenRAPM now has "Wins over Starter Average" Why: When thinking about MVP, who cares about 'replacement player'? How: Wins Added vs the avg starter this season - reducing minute weight during garbage time, & accounting for player effect on pace openrapm.com

the Mahdi is too humble to say he is worth $2 trillion even more reason to know that he is!
Claude thinks fair market cap for publicly traded Anthropic is somewhere between $500B and $800B btw

go on, tell me is this legit generalization for once, or did Anthropic and OpenAI buy a dataset from professional shadow puppeteers with Ph.Ds in projective geometry at 3x their hourly rates?
Introducing ShadowBench: can frontier models recreate shadow puppets? The model gets an RL environment with a curtain and 3D hands, and must pose them joint by joint until it matches a diagram from a 150 year old (!) puppet book. Crafted at @perpetual_si


Oh my goodness, it's the VS breakpoint
The average US state

There are six players available. Someone go find the other Caitlin.
Hey @C2_Cooper, you gonna suit up tonight?

Lots of followers out there grinding NFL film today. Reminder: if you see a pancake, share the info with us, as we're building a complete list across Week 8
The pancake block - when an offensive player flattens his opponent - has long lived as an unofficial football stat With more granular data, our NFL group has begun measuring pancakes. This weekend, we'd like your help Introducing: Pancake Week 1/4