NBA Player and Shooting Analysis

This project served as my introduction to dashboards with Jupyter Notebook. The final NBA dashboard is constructed in nba_jupyter_dash.ipynb. The other notebooks are my own personal guides to my widget and model building. In particular, widgets_presentations.ipynb, was my first take at widget creation. It is an interactive jupyter notebook where the user is able to specify a team, player, and game. Once specified, total shots made each game by the selected player will be presented as a bar graph. A separate line graph will also be displayed examining the percentage of shots made each quarter during that selected game.

The second jupyter notebook, shooting_patterns.ipynb, has an exploratory data analysis approach focused on modeling player shooting styles. Non-negative matrix factorization and its corresponding base vectors are used to determine player shooting styles. After which, a correlation matrix is implemented to determine similarities and differences between the players themselves, and finally, hierarchical clustering is used to group all players (of the selected season -- see below) into different "player type" groups. These two notebooks work independently of each other. Both are combined and used in nba_jupyter_dash.ipynb.

Requirements

You will need to have:

• Jupyter Notebook
• helper_basketball.py (provided above)
• wget installed (if not already working on linux machine)
• Jupyter Dashboards Layout Extension downloaded (for dashboard view)

Season Setup

Only applicable with widgets_presentations.ipynb and shooting_patterns.ipynb!

Some variables to consider changing are current_year and latest_season. Note that both parts of the project were created under 2017 current year and 2016-17 latest season. Code and results after the "Base Vector Experimentation" section in shooting_patterns.ipynb will not work when changing these variables (originally only for presentation), but can be easily replicated following the original code.

# choose season and year
# note: current_year is 20BB when latest_season is 20AA-BB
current_year = 2017
latest_season = '2016-17'

Note: in shooting_patterns.ipynb, the shots retrieval process is long and time consuming, and so when running it initially, a pickle file is created. Since this project was centered around the 2016-17 season, a pickle file of this year is provided above -- see allshots2016-17.pkl.
For seasons where no pickle file has been created, be sure to uncomment and run the cell that starts:

# run me if no pickle file for selected season
shots_params = {'PlayerID':'201939',
                'PlayerPosition':'',
                'Season': h.latest_season,

Running the Code

To run the setup and analysis parts simply download widgets_presentations.ipynb and/or shooting_patterns.ipynb and run via Jupyter Notebook.
To run the dashboard download nba_jupyter_dash.ipynb and run the file. Switch to dashboard view (possible with this download), and the dashboard will be ready for your use for however long and however many selections. Enjoy :)

Future work

In the future, I would like for get_nba_data() to not rely on wget function.
I would also like to find an active Jupyter Dashboards Server to deploy to.

References

For parameter specifications see Endpoint Documentation.
For shots data retrieval process in shooting_patterns.ipynb refer to arXiv:1401.0942.
Special thanks to Professor Sang-Yun Oh at UCSB for definitions work in helper_basketball.py and guidance throughout the project.