Views
No views yet
| Metric | Score | Description |
|---|---|---|
| Stability Score | 0.67701 | Official competition metric (Gini with stability penalty). |
| AUC | 0.8308 | Raw predictive power (Single Fold). |
| Slope | ~0.00 | Performance degradation over time (near zero is ideal). |
1.
2├── models/ # Pretrained CatBoost models (10GB+)
3│ ├── catboost_fold_1.cbm
4│ └── ...
5├── src/ # Core source code
6│ ├── data/ # Polars-based data pipeline & aggregation logic
7│ ├── features/ # Feature engineering & adversarial selection
8│ ├── models/ # Trainer wrapper for CatBoost/LGBM
9│ └── validation/ # Stability-aware cross-validation splitters
10├── notebooks/ # Experimentation labs (Jupyter)
11│ ├── 01_baseline... # Initial feasibility study
12│ ├── 02_feature... # Deep feature engineering (Depth 0/1/2)
13│ ├── 05_champion... # FINAL Training script (GPU required)
14│ └── ...
15├── docs/ # Detailed technical reports
16│ └── reports/ # Technical evolution, summary, and appendices
17├── training_artifacts/ # Logs and OOF predictions
18└── verify_model.py # Quick inference verification script1# Install Git LFS first
2git lfs install
3
4# Clone repository
5git clone https://huggingface.co/Lyes930/home-credit-risk-model-v1
6cd home-credit-risk-model-v1
7
8# Install Python dependencies
9pip install -r requirements.txtcsv_files or parquet_files folders inside a data/ directory in the root of this repo.1data/
2 ├── parquet_files/
3 │ ├── train/
4 │ └── test/
5 └── feature_definitions.csvpython verify_model.pynotebooks/05_champion_optimization.ipynb.data/ directory is populated.Payment -> Contract -> User) to capture deep historical credit behavior.1@misc{home-credit-risk-v1,
2 author = {Lyes930},
3 title = {Home Credit - Credit Risk Model Stability Solution},
4 year = {2025},
5 publisher = {Hugging Face},
6 journal = {Hugging Face Model Hub},
7 howpublished = {\url{https://huggingface.co/Lyes930/home-credit-risk-model-v1}}
8}