College Football Prediction App with Chatbot:

App Link:

https://jsandy.shinyapps.io/cfbapp/

Blog Post

https://garbage-time.github.io/posts/cfb-app/

Current Version Notes:

• XGB Model trained through Week 7.
• Removed Vegas features from predictive model
• Added SP+ features to model
• Added tracking table which displays teams in which the model predicts the correct cover outcome the most.

Note

The Chatbot can only access data pulled on teams prompted by the user's query. If you ask the Chatbot information like "Who is Duke's head coach", it will give you an outdated answer.

Purpose:

This project acts as a start-to-finish pipeline of securing college football data from multiple sources, training an XGBoost model for point differential predictions, and creating an app which users can intuitively experiment with the model.

Data Ingestion Sources:

• CBS Sports' complete team rankings webpage.
• cfbfastR API

Usage:

Execute 00 runner.R to pull/scrape data and train the models. Data will be saved to both the downstream folder and the cfbapp folder. Once this is complete, you can build the Shiny app within app.R

Important

This repo will not work without an R script called req.R, which should be created by external users and placed in cfbapp/helpers/. This script must then contain variables:

• OPENAI_API_KEY: String. The user's OpenAI api key.
• PROMPT2: A string dictating the rules for the LLM's behavior ('PROMPT2' because that's how it's referenced in other scripts).
• CFBD_API_KEY: String. The user's cfbfastR api key. This is needed for the schedule table. Further, in this script, the user must supply the following code:
  • Sys.setenv(CFBD_API_KEY = CFBD_API_KEY)
• DO NOT PUBLISH THIS UNLESS PRIVATE

Other scripts

Any scripts located in aws and misc are not required for the application to run. Scripts in misc are exploratory and pull data from different sources on the web, but aren't used in the model nor app. The script in the aws folder pushes the data created by 00 runner.R to an AWS S3 bucket called "cfbapp23". For this to work, you need to make sure you're signed into AWS via it's CLI on your local device.

Coming soon:

• Flowchart for ths README.md

Coming later:

• Experimentation with non-tree based models and potential inclusion.