kaggle-hill-climbing - Hill Climbing Ensemble
Build an ensemble using greedy forward selection, CuPy vectorized metric computation, and SciPy weight optimization.
Tags
Updated: 2026-09-20Capabilities
Typical Inputs
Typical Outputs
What this skill does
- Load out-of-fold predictions
- Evaluate individual model performance
- Perform greedy forward selection
- Optimize ensemble weights with SciPy
- Generate blended test predictions
- Log ensemble experiment results
Inputs
- Out-of-fold numpy prediction files
- Target ground truth labels
- Test set numpy prediction files
Outputs
- ensemble/hill_climb_test_pred.npy file
- experiments/hill_climbing.json file
Requirements
- Bash and Read tools permission
- Python environment with numpy pandas scipy scikit-learn
- Out-of-fold predictions from at least 3 models