This is the capstone project I completed as part of the BrainStation London Data Science bootcamp.
The aim of the project is to research whether we can predict how a football player may perform in a new team, as part of a transfer.
| File | Description |
|---|---|
| Notebook 1 - Data and Tools | This notebook is a simple exploration of the data and the tools provided by the SoccerAction Library. In this notebook I explore the actions of single match in the data set. |
| Notebook 2 - Create main dataset | Create new features, including pitch zones and details of the previous 5 actions. Includes code for generating the train and test datasets. |
| Notebook 3 - Initial Analysis | Using the training set, I perform an exploration of the data across all teams, focusing on end zone and actions performed in the context of xT and VAEP. |
| Notebook 4 - Baseline Models and Modelling Approach | This includes the pre-processing setup, and the baseline modelling iterations for each of the modelling approaches tested. Here I narrow down which models to take into the optimisation phase. |
| Notebook 5 - Model Selection and Hyper-parameter optimisation | This includes the iterations for reaching a final model selection, include model evaluation and explainability. Include GridSearches - Please note running this notebook can take hours (6-8 hours) |
| Notebook 6 - Final Model Analysis | Applying the final models to the project problem and analysing the results. |
| load_data.py | Simple util file to help load data between notebooks - used from Notebook 3 onwards. Can load data from s3 bucket or local. |
| pre_processing_utils.py | A set of helper functions to generate test and train datasets and to help configure the column transformers in the ML pipelines and GridSearches. |
| config.py | Contains the s3 URLs used in the load_data.py file - part of the .gitignore list |
| Capstone Project Report.pdf | Final project summary report |