language:
- en
tags:
- basketball
- nba
- sports
- tracking
pretty_name: 2015-2016 Raw Tracking Data from SportVU
source_datasets: https://github.com/linouk23/NBA-Player-Movements
2015-2016 Raw Tracking Data from SportVU
The modern era of basketball is characterized by the use of data to analyze performance and make decisions both on and off the court. Using tracking data combined with traditional play-by-play can allow for expert analysis of games.
Dataset Details
Dataset Descriptions
Tracking data is the finest level of basketball data, whereas play-by-play and box score data are also used. This dataset gives raw SportVU tracking data from each game of the 2015-2016 NBA season merged with play-by-play data. 2015-16 was the last season with publically available tracking data. This data has the coordinates of all players at all moments of the game, for each game in the season. There is also more information such as descriptors for players on the team (and their unique IDs) and the teams playing (and their unique IDs). Further, descriptors of the play that occured at each event is present, and the team in possession during the event, along with more necessary features.
- Collected By: SportVU, Basketball Referece
- Shared By: Kostya Linou, Dzmitryi Linou, Martijn De Boer, Sumitro Datta
Dataset Source
- Repositories:
Uses
This dataset has many potential uses. Primarily, visualization of plays, as illustrated in the initial repository is possible, creating a comprehensive view for analyzing actions on court. Beyond that, models could be trained to recognize certain play types or actions, as illustrated in previous papers (see Stephanos et al., 2022). Analysis of defensive control could be performed by examining the data spatially. Even further, a broadcast tracking model could be creater if video data could be obtained and connected to each moment of collection. This would create a model where video frames are mapped to tracked coordinates, increasing the accessibility of tracking data as only publically available video footage is necessary.
- Stephanos et al.: https://www.sloansportsconference.com/research-papers/using-hex-maps-to-classify-cluster-dribble-hand-off-variants-in-the-nba
Dataset Structure
The data is in the format of a dictionary:
- 'gameid': str
- 'gamedate': str
- 'event_info':
- 'eventid': str
- 'type': int
- 'possession_team_id': float
- 'desc_home': str
- 'desc_away': str
- 'primary_info':
- 'team': str
- 'player_id': float
- 'team_id': float
- 'secondary_info': same format as primary info
- 'visitor':
- 'name': str
- 'teamid': int,
- 'abbreviation': str
- 'players': list of the dictionaries in the form of the following
- 'lastname': str
- 'firstname': str
- 'playerid': str
- 'number': int
- 'position': str
- 'home': same format as visitor
- 'quarter': int
- 'game_clock': float
- 'shot_clock': float
- 'ball_coordinates':
- 'x': float
- 'y': float
- 'z': float
- 'player_coordinates': list of the dictionaries in the form of the following
- 'teamid': int
- 'playerid': int
- 'x': float
- 'y': float
- 'z': float
Requirements
To load the data, you must run
import py7zr
Dataset Creation
Curation Rationale
The reason for uploading this data to huggingface, is that in its current .7z form, the data is less accessible, and requires unzipping many files and then combining to access. Also, more sources for easily accessible tracking data, even if also available elsewhere, increase the chances of long-term preservation and accessibility for future NBA fans.
On top of that, tracking data combined with play-by-play data is ideal format of sports data, as there is little confusion and allows for better labeling of events.
Source Data
From creator StatsPerform, "the SportVU camera system is installed in basketball arenas to track the real-time positions of players and the ball at 25 times per second." These methods were used to capture the data in this dataset.
Bias, Risks, and Limitations
Since this data is not up-to-date, and the tracking data for the last eight seasons is private and unreleased, the continued spread of this specific data may not be representative of the current state of NBA tracking data. Thus, users that learn how to manipulate it may not be adequately prepared for work in basketball organizations. Further, analyses performed on the dataset may not be reflective of the current state of professional basketball. However, since this was the last iteration of publicly available tracking data, I believe increasing its availability is important.
Dataset Card Author
Donald Cayton; [email protected]