Tuzu Model

A set of scripts to create point predictions for Fantasy Premier League, based on the great repository:
vaastav/Fantasy-Premier-League.

To ensure accurate predictions, the models are trained using only matches where players played at least 60 minutes. Then future predictions are generated based on model results, xMins and availability.

I plan to improve the model in the future, and I love discussing FPL and FPL prediction models.
Feel free to reach out to me at szymone25@gmail.com.

A frontend app for this project is available here:
minifpl-front
You are welcome to both of these and adapt them to your needs.

---

How to use

Run the notebooks manually

1. Clone the repository and navigate to the project folder:

   git clone https://github.com/FantaInz/tuzuModel.git
   cd tuzuModel

2. Set up a virtual environment:

   bash setup_venv.sh <name_of_env>
   source <name_of_env>/bin/activate

3. Start Jupyter Notebook:

   jupyter notebook

Run everything automatically

1. Clone the repository and navigate to the project folder:

   git clone https://github.com/FantaInz/tuzuModel.git
   cd tuzuModel

2. Set up a virtual environment:

   bash setup_venv.sh <name_of_env>
   source <name_of_env>/bin/activate

3. Run the entire workflow (with optional database saving):
   If you want to save predictions to the database, set the appropriate database variables before running the script:

   export DB_NAME=my_database
   export DB_USER=my_user
   export DB_HOST=my_host
   export DB_PASSWORD=my_password
   export PORT=your_port

   Then run the script:

   RUN_SAVE_TO_DB=true bash execute_whole.sh  # Set to true if saving to DB, otherwise omit RUN_SAVE_TO_DB

Run in Docker

1. Download and install Docker:
   Docker Installation Guide

2. Clone the repository and navigate to the project folder:

   git clone https://github.com/FantaInz/tuzuModel.git
   cd tuzuModel

3. Build the Docker image:

   docker build -t <app_name> .

4. Run the container:

   • Without saving predictions to the database:

     docker run --rm <app_name>

   • With saving predictions to the database:
     Make sure to set the correct database environment variables:

     docker run --rm \
         -e DB_NAME=my_database \
         -e DB_USER=my_user \
         -e DB_HOST=my_host \
         -e DB_PASSWORD=my_password \
         -e PORT=your_port \
         -e RUN_SAVE_TO_DB=true \
         <app_name>

---

Folder Structure

• predictions/ – Contains predictions for upcoming gameweeks:

  • -raw → Predictions without xMins adjustments.
  • no-availability → Predictions with xMins but without considering player availability.

• models/ – Contains trained models.

• created_csv/ – Contains mappings between FPL and Understat data, as well as player names across seasons.

---

Flow

Step	Notebook	Description
1	repo_cleanup.ipynb	Retrieves data and deletes unnecessary files and folders.
2	entry_processing.ipynb	Preprocesses the data to prepare it for training and predictions.
3	gk_training.ipynb	Trains a CatBoost model for goalkeepers.
4	def_training.ipynb	Trains a Random Forest model for defenders.
5	mid_training.ipynb	Trains a CatBoost model for midfielders.
6	fwd_training.ipynb	Trains a CatBoost model for forwards.
7	get_prediction_data.ipynb	Calculates the data needed for predicting the next gameweeks.
8	predictions.ipynb	Generates predictions with and without xMins adjustments.
9	add_availability.ipynb	Fetches player availability from the FPL API and adjusts predictions.
10	save_to_database.ipynb	Saves predictions to the database (used in BackEnd).