iiinova — where your digital self keeps evolving

Nathan Walker @bbstats.bsky.social · 28/09/2026

there are now 5 good AI models: Opus 5.5 Medium thru Max Sonnet 5.5 xHigh

000

Nathan Walker @bbstats.bsky.social · 02/09/2026

faster than Sol 5.6 high as well

000

Nathan Walker @bbstats.bsky.social · 22/08/2026

me to claude "pretend you are a brilliant communicator who loves ensuring folks QUICKLY and clearly understand your intent:"

000

Nathan Walker @bbstats.bsky.social · 04/08/2026

Hoping to make ChimeraBoost the tool of choice for quick R&D of data science projects github.com/bbstats/chim... My favorite part is the 'quality' parameter - you can very simply and understandably adjust the quality/speed of the model with a single parameter!

001

Nathan Walker @bbstats.bsky.social · 03/08/2026

oh yeah! how's the weather life?

110

Nathan Walker @bbstats.bsky.social · 01/08/2026

Brier Skill Score for president

110

Nathan Walker @bbstats.bsky.social · 28/07/2026

Introducing 'replay'! github.com/bbstats/chim... 1. Fit with early stopping 2. Using the same trees & splits (skipping ~85% of train time), re-calculate the leaf values (& other data-dependent parameters). In my testing, 10-40% speed improvement over a 'full-refit' and the speed scales with size!

github.com

Reuse the tree structures when refitting on the full data: same accuracy, a third less fit time by bbstats · Pull Request #41 · bbstats/chimeraboost

What changes: refit_full now defaults to "replay". Same accuracy, roughly a third less time to fit. Why we currently fit twice With early_stopping=True (the default), ChimeraBoost holds o...

000

Nathan Walker @bbstats.bsky.social · 28/07/2026

3. The simplest solution is to do a full refit! Something I have done manually in the past. But this roughly doubles the training time of the model What if there was some way to retain most of the tree structure your GBDT, but increase the data it sees to the full dataset?

100

Reposted by Nathan Walker

Nathan Walker @bbstats.bsky.social · 25/06/2026

posting this for no reason whatsoever TSA per 100 vs Adj PPS (TS%x2) League average Adj PPS of 1.162

132

Nathan Walker @bbstats.bsky.social · 25/06/2026

posting this for no reason whatsoever TSA per 100 vs Adj PPS (TS%x2) League average Adj PPS of 1.162

132

Nathan Walker @bbstats.bsky.social · 24/06/2026

loved the last clip in particular

020

Nathan Walker @bbstats.bsky.social · 16/06/2026

Doing the minimal possible SEO here but - I (and Claude) developed a python-only (e.g. numba!) backend Gradient Boosting library that is insanely fast on small/medium data (eg your prototyping phase!) pip install it and tell me what you think github.com/bbstats/chim...

github.com

GitHub - bbstats/chimeraboost: A Python-backend gradient boosting library

A Python-backend gradient boosting library. Contribute to bbstats/chimeraboost development by creating an account on GitHub.

030

Reposted by Nathan Walker

Basketball Poetry @basketballpoetry.com · 04/06/2026

Karl-Anthony Towns was never an underdog. Finally, however, he’s no longer underappreciated. More here: www.basketballpoetry.com/p/karl-antho...

basketballpoetry.com

Karl-Anthony Towns was never an underdog

Karl-Anthony Towns has never been an underdog.

071

Reposted by Nathan Walker

Jon Bois @jonbois.bsky.social · 03/06/2026

codes are code coded

1654353

Reposted by Nathan Walker

Dave DuFour 🐉 @davedufour.bsky.social · 01/06/2026

Victor Wembanyama is basketball Genghis Khan

3503

Nathan Walker @bbstats.bsky.social · 31/05/2026

when the why is also being mentioned all the time, in separate opining

010

Nathan Walker @bbstats.bsky.social · 31/05/2026

Lots of opining about Thunder players performing worse, not a lot of folks mentioning WHY

230

Nathan Walker @bbstats.bsky.social · 30/05/2026

I'm not 100% sure what you mean but I do think there are probably hundreds of caveats, just named the first two that come to mind

000

Nathan Walker @bbstats.bsky.social · 30/05/2026

Caveat A: Specialists for specialist scenarios (single defensive possession stoppers in close games) probably have more utility there for example Caveat B: "Over the course of their contract" of course depends on how tradeable the contract is / becomes.

010

Nathan Walker @bbstats.bsky.social · 30/05/2026

Players shouldn't be evaluated from their possession-level or play-level ability, but by their ability over the course of their contract.

230

Nathan Walker @bbstats.bsky.social · 26/05/2026

GPT 5.1 1455 Arena Elo (current) November 12, 2025 $7.81 per million weighted tokens Gemma 4 1451 Arena Elo (current) April 2, 2026 $0.34 per million weighted tokens

010

Reposted by Nathan Walker

ArtButMakeItSports @artbutmakeitsports.bsky.social · 19/05/2026

Angels Appearing before the Shepherds, by Henry Ossawa Tanner, 1910

162499431

Nathan Walker @bbstats.bsky.social · 14/05/2026

at a previous employer our acronym was certainly adjacent to this

010

Nathan Walker @bbstats.bsky.social · 11/05/2026

Adam Silver doesn't know I have kid dropoff at 730am

010

Nathan Walker @bbstats.bsky.social · 05/05/2026

Turnovers Reduced (TR) per-100 in the playoffs for remaining players, Top and Bottom 10 Uses AST/2PA/3PA/FTA to predict TOV (XTOV). Would be better using Ast Opps especially w/ smaller sample size but, still useful. data from @basketball-reference.com

121

Nathan Walker @bbstats.bsky.social · 04/05/2026

I've softened slightly on my TLJ love - I think there are some clear anti-fan tendencies in it that I think were unwise at least. if ROTS could have convinced me of Anakin's turn or that he was friends with Obi-Wan it would be higher on my list.

110

Nathan Walker @bbstats.bsky.social · 03/05/2026

Boston is Harden-ing like Kakuna send tweet

020