ABOUT
SOMEDAYGITHUB
HOOPS
LAB
LOADING SEASON DATA...
INSERT COIN
● HOOPS LAB1P  HI 999999  CREDIT 01

HOOPS
LAB

BUILDING DATA SCIENCE SKILLS
THROUGH MY PASSION FOR THE NBA

▶ PRESS START
VIEW PROJECTSGITHUB / MEZZ2
2
PTS 000
▼ SCROLL
The work

FIVE PROJECTS,
ONE DATA STACK.

Each project builds on the last — shot quality feeds game prediction, which feeds trade value. Start with the foundation and watch the stack compound.

01

Shot Quality Model

xFG
In Progress

How much better or worse is a player shooting than expected, given shot difficulty?

An expected field goal (xFG) model that accounts for shot distance, zone, type, game context and defender distance. Output: a calibrated probability for every shot, plus a leaderboard of over- and under-performing shooters.

Model
Logistic → Gradient Boosting
Metric
Log loss + calibration curve
Target
4 weeks to v1
NotebookWrite-up · coming soon
SHOT CHART
FG% vs league avg
by location
COLDHOT
Illustrative — live model output coming soon
The roadmap · 02–05
02

Game Prediction Engine

Planned

Can we predict game outcomes better than Vegas using team-level efficiency metrics?

Builds directly on the xFG model — team shot quality becomes an input feature. Explores how much pregame data (pace, offensive / defensive rating, rest days, travel) predicts the final score.

Builds on  01 · Shot Quality
Inputs  Pace, ORtg/DRtg, rest, travel
03

Trade Value Predictor

Planned

What is a player actually worth in a trade, quantitatively?

Combines on-court production (informed by the shot-quality work), contract data, age curves and positional scarcity into a single tradeable value score. Inspired by Basketball-Reference WAR — but built from scratch.

Builds on  01 · 03 production signal
Approach  WAR-style, from scratch
04

Play-by-Play Clustering

Planned

What types of possessions actually exist, and which teams and players run them most?

Unsupervised clustering on play-by-play sequences to surface possession archetypes — pick-and-roll, isolation, transition. Feeds naturally into both trade value and game prediction.

Method  Unsupervised clustering
Feeds  02 · 03
05

Draft Class Projection

Planned

Which college and international prospects translate to NBA production, and why?

Historical draft-class analysis plus a projection model. The capstone — it requires the full data infrastructure built across projects 1 through 4.

Requires  Infra from 01–04
Type  Projection model
Backlog

OTHER QUESTIONS WORTH CHASING

Pulled out during planning — they didn’t fit neatly into the five projects, but they’re too good to lose.

Binary shot prediction (made / missed) as a standalone classification benchmark
Player shot-selection ranking — who takes the highest-quality shots, regardless of outcome?
Does shot quality predict future performance? (regression to the mean)
Game-context effects on shooting — how does pressure change shot quality?