Disc golf score calculator:
What This Does: provides weather-based prediction of scores for league disc golfers at three San Francisco Bay Area disc golf courses (Golden Gate Park, Lake Chabot, and Berkeley Aquatic Park).
Why We Did It: this is a fun way to see how machine learning works. This project makes use of scikit-learn's ElasticNetCV, with a customized way of balancing the bias-variance tradeoff. In particular, because the model is fully interpretable, we can directly evaluate the realism of the predicted model parameters and reject overfitted solutions.
Where Is This Deployed? The user interface is currently deployed on Heroku at disc-golf-deep.herokuapp.com. To use the app, the only requirement is a modern browser (tested on Chrome 75 for Windows 10 and Chrome for MacOS) with JavaScript enabled.
The main repo is on GitHub at https://github.com/discgolfpredictiveinsightsteam/Main-Repo
The repo has two parts, "scoreCalculator" contains all of the code for the deployed app, while "models" contains all of the code needed to train and generate the models. Note that the model generation software (mainly Jupyter notebooks) will make a local copy of a csv file named "score_model.csv" -- to update the model, this file needs to be moved to the "data" directory in "scoreModel" before deploying the app.
The base score data is provided in a csv file called "scores.csv" in the "data" directory.
This file is only needed in the event that the sqlite database needs to be re-generated.
Weather data is provided by Dark Skies through its "time machine" API. If you wish to
test this functionality outside of the app, you will need to supply your own API key
via system environment variables (the variable names can be seen in app.py, we currently
use two variables supplied with the same API key for future development purposes).
The "data" directory also contains a set of predicted and actual scores for validation purposes.
The "models" directory contains the workflow for training the models. Standard processes
are contained in the "notebooks" folder, including the three latest models, with titles
indicating their vintage. Only the latest is needed, but the previous two are available
for rollback or comparison purposes. THe folders "eda" and "validate" contains additional
exploratory and validation work.
Each notebook contains a self-explanatory introduction that lists its requirements and file structure.
Within the "models" folder, there is a mirror of the "static" folder for the app. This folder is provided as a convenient archive copy for testing purposes. Because it is updated infrequently, it is not guaranteed to be up-to-date with the latest app version.
Acknowledgements:
We'd like to thank the Golden Gate, Lake Chabot, and East Bay disc golf leagues for sharing their public score data.
Lastly ...
Here is a cat for Andi.
