Soccer data analysis and predictions

About

This is github repository for project for "Machine learning" class at FIB UPC.

Team

• Martin Galajda
• Florian König
• Aitor Ortiz de Latierro

Guidelines working with the projects

• ⚠️ When working with R scripts, working directory is set to the project root (all code should follow this convention! 🙏)

Directory structure

models/

• The purpose of this directory is to contain models which will be built for different predictions (match result, goals scored, player analytics)

• models/match_result

  • predicting match result (win, tie, loss)

• models/goals_scored

  • predicting goals scored

loaders/

• The purpose of this directory is to define functions which will load different data with different features (used in models). Possibly uses precomputed features which take a bit time to compute.

feature_extraction/

• The purpose of this directory is to define everything related to feature extraction.

• precompute_features/

  • this directory contains functions which when executed (source(path) in R) will make sure that cached precomputed features are present in data/ directory
  • this are used by some loaders to make sure that precomputed features are saved in csv file in data/ directory

  • precompute_attack_strength_ratios.R

  • compute attack strength ratio for each match by taking into account all matches from past (so we don't leak future information into our data)
  • saves precomputed features in data/ directory

  • precompute_win_ratio.R

  • compute win ratios for given match
  • takes parameter which determines how many matches from past to take into account
  • saves precomputed features in data/ directory

data/

• The purpose of this directory is to contain all data for predicting / analysing
• Specifically, we store here even precomputed features which are not easily computed on the run (take ~ 5 min)

data_conversion/

• The purpose of this directory is to contain all code that extracts usable data from the SQLite database.

validation/

• The purpose of this directory is to contain all code related to validation and selecting right models/features. Some of the scripts compute CV-errors for different subset of features/ parameters and save them as Rdata (because it takes a lot of time to compute).

results/

• The purpose of this directory is to observe obtained results (for match_result). It uses saved RData to present obtained results from cross-validation and test error from best model.

How to run

Requirements

• R environment (RStudio preferably)

Steps

• set root of the project as working directory (some scripts require that)
• run the code

Interesting scripts to run

• data_conversion\X.R generates necessary data\db_X.CSV database from original data/database.sqlite.
• validation/match_result/*.R - scripts contain logic for feature selection, tuning parameters for different models. However, execution time is very big for some scripts (therefore they store results in results/ directory after computing).
• results/match_results/validation/*.R - scripts contain interpretation of results achieved by running scripts from validation/match_result.R scripts. They use saved RData from results/match_results/saved_Rdata directory.
• results/match_result/test_error_best_model.R - script computing test error for best model for predicting match results
• results/match_result/compare_models.R - script comparing different models for predicting match result and observing their respective CV-errors.
• analyze_players.R - full code for Player Analysis section
• models/goals_scored/*.R - self-contained scripts to tune and train the different number of goals models including the baseline + goals_distribution.R plot script.