Awesome

NBA Shots DB

Rails app to populate a PostgreSQL database containing every shot attempted in the NBA since 1996.

Blog post with some analysis of the data: https://toddwschneider.com/posts/nba-vs-ncaa-basketball-shooting-performance/

Data comes from the NBA Stats API. As of March 2018, the database contains ~4.5 million shots from 2,000 players, and takes up 1.5 GB disk space.

NCAA men's college basketball data

NCAA data is available via Sportradar on Google BigQuery. The mbb_pbp_sr table contains men's basketball shot chart data since 2013.

The app/lib/ncaa_shots subfolder of this repo contains additional scripts to process the NCAA shots data and merge it with the NBA data, see here for more info.

Setup

Assumes you have Ruby and PostgreSQL installed

git clone git@github.com:toddwschneider/nba-shots-db.git
cd nba-shots-db/
bundle exec rake db:setup

Usage

Import all available data (will take a few hours):

bundle exec rake import_all_shots
bundle exec rake jobs:work

Alternatively, import data selectively by going into the Rails console and running, e.g.:

lebron = Player.find_by(display_name: "LeBron James")
lebron.create_shots(season: "2017-18", season_type: "Regular Season")
Delayed::Worker.new.work_off

Notes

player#create_shots uses Postgres's COPY command instead of Rails's #create method because COPY is ~10x faster
Shots have no natural unique identifier: no external IDs, no guarantee that there's only 1 shot from 1 player at the same second of the same game, etc. Accordingly, player#create_shots deletes and replaces data every time it is run.

Closest defender aggregates [Deprecated]

The original version of this project had a table called closest_defender_aggregates that included player/season aggregates segmented by shot distance and the distance of the closest defender at the time of the shot.

Unfortunately, as of December 2020 (and perhaps earlier), it seems like the NBA API no longer makes this data available. The NbaStatsClient#shot_stats_by_closest_defender method currently only seems to work if the shot_dist_range argument is empty. This means you can get a player's shooting performance segmented by closest defender distance, but you cannot further segment by shot distance.