DEFCON

This repository contains a full machine-learning pipeline for training a model, generating predictions, and producing a submission.csv file for Kaggle.

1. Requirements

• Python 3.9+
• Git
• A terminal (Linux, macOS, or Windows PowerShell)

2. Clone the repository

git clone https://github.com/simy46/DEFCON.git
cd DEFCON

3. Venv

Linux / macOS

python3 -m venv venv
source venv/bin/activate

Windows

python -m venv venv
venv\Scripts\Activate

4. Dependencies

pip install -r requirements.txt

5. Config file structure

Each config file has three parts: model, preprocessing, and hyperoptimization.

5.1. model

Defines models base hyperparameters.
When running hyperopt:

• parameters set to None fall back to the values here
• the final *_best.yaml overwrites this section with optimized values

5.2. preprocessing

Controls all preprocessing steps.
Each step has enabled: true/false.
Only steps with enabled: true are applied.
This structure is identical for every model.

5.3. hyperoptimization

Enables hyperparameter search (grid, random, or optuna).
Only parameters defined here are searched.
Parameters set to None keep their value from the model section.

1. Running the main pipeline

The pipeline is controlled through a YAML config file.

python main.py --config config/model_rt_best.yaml

Other available configs are listed on /config folder. The command that we gave is to run the same configs as our submission.

2. What the script does

1. Loads the YAML config
2. Loads training and test data
3. Applies preprocessing (scaling, PCA, etc.)
4. Trains the model defined in the config
5. Generates predictions on the test set
6. Saves a submission file in submissions/submission_.csv

3. Output

submissions/submission_2025-01-04_14-32-10.csv

1. Run hyperparameter search

This script searches for the best hyperparameters and generates a new optimized config file.

python find_best_params.py --config config/model_<name>_best.yaml

Other available configs are listed on /config folder.

2. What the script does

1. Loads the YAML config
2. Loads training and test data
3. Applies preprocessing (same pipeline as the main script)
4. Builds the model and its search space
5. Runs hyperparameter search (Grid Search, Random Search, or Optuna)
6. Logs the best score and best parameters
7. Writes a new file: config/model_rt_best.yaml with the updated hyperparameters

3. Output

config/model_<name>_best.yaml

After generating this file, use it on the main.py pipeline

python main.py --config config/model_<name>_best.yaml