yashkarle / data-science-notes Goto Github PK

TL;DR

List of approaches for building a playlist recommender using Spotify data

Article Link

https://github.com/mrthlinh/Spotify-Playlist-Recommender#proposed-solutions

Author

Linh Truong
http://mrthlinh.github.io/

Key Takeaways

• Collaborative Filtering
• Approaching a ML problem

Comments/ Questions

TL;DR

List of ideas, articles and code snippets to understand the use of a Poisson model to predict goals and matches

Blog Posts

Poisson distribution R tutorial
Danny Page expected goals simulator

Project Idea

• Construct a Poisson model from "number of shots + xG of those shots" predicting "goals" for a given team
• a) Goal simulation (xG, xGA, shots_for, shots_faced_against; clean-sheet probabilities can be included here too)
• b) Match simulation (win_t1/draw/win_t2 probabilities; worth considering opposition strength (FDR from FPL))
• c) Entire season simulation (over all 38 league matches; possibly take into account form over last n matches)

Key theoretical takeaways

• Randomness in Football in general
• Poisson counting process, Poisson distribution
• Simulation of PL matches based on shots (quantity and quality)

Useful Code Snippets

R code snippets

h_scored = rpois(1, 1/2 * (t_ave_h_s + t_ave_a_c))
a_scored = rpois(1, 1/2 * (t_ave_a_s + t_ave_h_c))

Comments/ Questions

TL;DR

List of ideas, articles and code snippets to understand ...

Blog Posts

https://theathletic.com/3044091/2022/01/02/the-five-kingdoms-of-football/
https://theathletic.com/3050603/2022/01/05/who-never-fouls-who-presses-who-hoofs-it-what-our-data-experiment-you-about-your-team/

Project Idea

Very simply, the idea is to rely on John Muller's analysis done for teams and extend it for players and coaches

Key theoretical takeaways

• UMAP - Uniform Manifold Approximation and Projection for dimension reduction
  https://umap-learn.readthedocs.io/en/latest/

• GMM - Gaussian Mixture Models
  https://scikit-learn.org/stable/modules/mixture.html

Comments/ Questions