https://github.com/sharmaabhishekk/interactive-freeze-frames-xg

Interactive site to mess around with xG values
https://github.com/sharmaabhishekk/interactive-freeze-frames-xg

football html-css-javascript tensorflow

Last synced: 4 months ago
JSON representation

Interactive site to mess around with xG values

• Host: GitHub
• URL: https://github.com/sharmaabhishekk/interactive-freeze-frames-xg
• Owner: sharmaabhishekk
• Created: 2021-09-11T18:34:32.000Z (almost 5 years ago)
• Default Branch: main
• Last Pushed: 2024-01-14T04:06:40.000Z (over 2 years ago)
• Last Synced: 2024-01-14T10:55:35.953Z (over 2 years ago)
• Topics: football, html-css-javascript, tensorflow
• Language: Jupyter Notebook
• Homepage: https://sharmaabhishekk.github.io/Interactive-freeze-frames-xg
• Size: 3.66 MB
• Stars: 6
• Watchers: 1
• Forks: 3
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# xG Board

This site is meant as a demo to go along with this blog post [xG with CNNs]("https://www.opengoalapp.com/xg-with-cnns-full-study"). The Expected Goals (xG) model developed there is a CNN-based model trained on Statsbomb's freeze frames dataset. The features it takes into account are:

* shot location

* location of the goalkeeper

* location of the defenders

## Model details

To learn more about the model, it is strongly suggested you read the full write-up linked above. The important part is the model architecture and the inputs.

For each shot scenario, the model takes into account three things - the location of the ball, the location of defenders, and the location of the goalkeeper, and merges them to form a single input.

![Model Inputs](https://raw.githubusercontent.com/sharmaabhishekk/Interactive-freeze-frames-xg/main/public/images/model-input.png)

The model itself has 3 [Convolution + Max-pooling] layers, followed by a single dense layer to flatten out the outputs from the convolution layer. Finally, we predict a single value which is the the probability of the shot resulting in a goal (AKA, the xG value).

![Model Architecture](https://raw.githubusercontent.com/sharmaabhishekk/Interactive-freeze-frames-xg/main/public/images/final_arch.png)