ensembling

What problem does it solve?

This Skill addresses the common machine learning challenge where individual models reach a performance plateau, enabling users to achieve higher scores by intelligently combining their predictions.

Core Features & Use Cases

• Prediction Combination: Integrates multiple model predictions using various techniques like weighted blending, rank averaging, stacking, and greedy hill-climbing.
• Model Diversity Assessment: Includes checks for pairwise OOF correlation to ensure ensemble members contribute unique information.
• Use Case: After training several diverse models (e.g., LightGBM, XGBoost, a Neural Network) for a Kaggle competition, use this Skill to blend their out-of-fold predictions to create a final submission that outperforms any single model.

Quick Start

Use the ensembling skill to combine the OOF predictions from 'lgbm' and 'xgb' models with their corresponding test predictions, optimizing weights using the provided training labels 'y_train'.