Skip to content

Latest commit

 

History

History
72 lines (44 loc) · 2.37 KB

readme.md

File metadata and controls

72 lines (44 loc) · 2.37 KB

Football Models and Visualisations


This repository is a personal project where I develop football models and visualisations.

It's build on top of different technologies such as:

  • Python: for data processing and machine learning (Tensorflow, Scikit-Learn and MLFlow).
  • R: all data-visualisation stuff (dplyr, ggplot2, magick).
  • Postgres SQL: storing and querying data easily.
  • Docker: to orchestrate all these elements together and easy install/startup.
  • Google Cloud Plateform: to run heavy jobs in cloud.

Samples & Results

pass-network

pass-sonar

assist-shot cluster map

rollmean

xg-map

xa-map

image

image

image

Architecture

The project contains five folders:

  • ./app : maybe deprecated design, mostly for data integration and easy of use.
  • ./data: where raw data are stored (in addition to the database). Not in git 😉.
  • ./model: machine learning models (expected goal for example).
  • ./src: source file for crawlers, database connection/ingestion, SQL queries, etc...
  • ./visualisation: source code for data-visualization, most recent works (on maps) are in ./visualisation/maps subfolders.

Usage

Running model applications

  1. Start corresponding container: docker-compose up -d model.
  2. For ease of use, going into the container: docker exec -it model bash
  3. All data are mapped to the local environment: cd /data
  • Expected goal model: python -m model.expected_goal.main --help.
  • Expected assist model: python -m model.expected_assist.main --help.
  • Possession2Vec model: python -m model.pass2vec.main --help

Running Passmaps vizualisations

  1. Start corresponding container: docker-compose up -d passmap.
  2. Go to http://localhost:8082/.

TODO

  • Improve models
  • Add more documentations.
  • Clean some viz stuff.
  • Better Docker management.

Contacts

Any questions/improves on Twitter @Ben8t.