PL Bets

Description

Summary

This repository contains the source code and documentation for the PL Bets project, a machine learning full-stack docker application written with in Python, and a frontend written in ReactJS.

The project uses neural network modeled with Tensorflow and Keras to predict the outcomes of a future match based on individual player statistics for each player in the squads.

Additionally, the frontend showcases outcome prediction for each of the matches in the upcoming gameweek, using logistic regression, based on the historic team results.

Structure

• Frontend container - Hosts web server
• Postgres container - Holds all player and team data
• Prediction container - Contains model and scripts to create and recreate model, and API to handle and serve predictions
• API container - Connecting frontend to database, serving player and team data

Table of Contents

• Data Sources
• Installation
• Usage
• Contributing

Sources

Data used in these models starts in the 2017-2018 season, bar the squad data, which is being fetched from the 2016-2017 season, onwards:

Player stats
Match facts
Fixtures
Squads

Installation

Dependencies

Ensure you have docker installed on your machine. If not, Docker can be downloaded here, selecting the correct download based on your operating system.

Usage

Makefile contains commands to setup and run the docker compose for this project

Enter app folder

cd app

Start project

make devstackbuild
make devstackup

Rebuild project

make devstackrebuild

Restart project

make devstackreboot

Stop

make devstackdown

See Makefile to see all commands available

After the containers have started, navigate to the frontend at localhost:3000

Contributing

Dominic Lawson

Notes

• Compatibility issues between PyPi, Tensorflow and h5py mean that h5py==3.9.0 & tensorflow==2.12.0 must be the versions used with these packages
• In VSCode add :

"python.analysis.extraPaths": [
    "./app/utils"
]

to your settings.json to suppress errors about the utility modules

TODO

• Use Exec/Eval (dynamic code execution) to allow analysis of data in the frontend

• Decorators for re-used functions

• Rearrange file structure for easier deployment with Terraform

  • Use volumes to mount common files/directories to containers
  • Use GitHub Actions to copy common files/directories into build context before running the build step

• Set up CI/CD pipeline for deployment to AWS

• Deploy to AWS with CDK/Terraform

• API-FOOTBALL.COM to get more data

• Manager page

• Scrape manager data - Wikipedia [x]

• • Typical stats per position for each manager
• • Typical overall team stats for each manager
• • Per 90 stats per manager
• • Sum team stats during managers career to identify key characteristics of each managers personal philosophy/tactics
• • Perform PCA to identify manager/tactic types
• • Identify key metrics for each tactic
• • Display in Tactics/Coaches page
• • Find some way to include this in outcome prediction
• • • Use model to determine effectiveness of each tactic against each other

• FPL page

• • Solver - Generate the best team based on fpl rules
• • Ratings for each player for each upcoming game
• • • expected points tally for each game
• • Compare players/positions/teams

• Provide choice stats for each game, comparing the two teams and tactics

• Caching

• General code cleanup and re-factoring

• Review "Separation of concerns"

• prediction model edits

• Visualisation Page

  • Add Managers to entities that can be visualised

Considerations

• To increase the number of leagues considered, teams, match_logs, managers, players data downloads will need to be significantly modified