FlightRank 2025: Aeroclub RecSys Cup — 4th Solution

This repository contains the source code for our best XGBoost model, achieving a LB of 0.52795 and PB of 0.53552.

Kaggle Write-Up

4th Solution: XGBoost Model

Environment Requirements

• Python Version: 3.11
• Memory: Running the full training pipeline (loading the dataset and feature engineering) requires approximately 400 GB of RAM
• GPU: Recommended for accelerating XGBoost model training

File Structure

├── extract.ipynb       # Extract additional information from raw JSON files
├── flight.ipynb        # Main notebook
├── requirements.txt    # Python dependencies
├── run.sh              # Script to run notebook in the background
└── src
    ├── __init__.py
    ├── data.py             # Data splitting
    ├── feature.py          # Feature engineering and selection
    ├── feature_specs.py    # Feature specifications used in the model
    ├── params.py           # Model hyperparameters
    ├── plot.py             # Visualization functions
    └── utils.py            # Utility functions: evaluation, reranking, prediction

How to Run the Code

1. Download the data from Kaggle Aeroclub RecSys 2025 Data and place it in the ./data/ directory. Follow the instructions to unzip jsons_raw.tar.kaggle.
2. Create a Conda environment and install the required packages from requirements.txt:

   conda create -n FlightRank python=3.11
   conda activate FlightRank
   pip install -r requirements.txt

3. Run extract.ipynb to extract additional information from the raw JSON files.
4. Configure flight.ipynb:
   • FULL = True — train on the full training dataset.
   • FULL = False — train on 90% of the training data for local validation.
5. Optionally, adjust the training-validation split sizes in utils.py:
   • TRAIN_VAL_SIZE — size of the training-validation split.
   • TRAIN_ALL_SIZE — size of the dataset when modifying the training set.
6. To add more features, modify or add functions in feature.py and include them in the feature_engineering workflow.
7. For feature selection, update the FeatureSpec class in feature_specs.py by adding or removing features as needed.
8. To use a trained model for predictions, go to the "Model Training" cell in flight.ipynb, uncomment the code that loads the model, and comment out the code that performs training.