⚡ LightGBM Flashcards
A gradient boosting framework that uses tree-based learning algorithms
🌳 What is LightGBM?
A high-performance gradient boosting framework optimized for speed and accuracy, especially on large datasets.
🚀 Speed Advantage
Uses histogram-based algorithms and leaf-wise tree growth, making it significantly faster than XGBoost in many cases.
📊 Dataset Compatibility
Best suited for large datasets with thousands of features. Handles both categorical and numerical features efficiently.
🧪 Use Cases
Fraud detection, click-through rate prediction, ranking tasks, and any tabular supervised learning problem.
⚙️ Core Features
Supports parallel learning, efficient memory usage, GPU training, and early stopping.
📦 Installation
Install with pip install lightgbm. Requires CMake and compilers for advanced builds.
📈 Tuning Parameters
Key parameters: num_leaves, learning_rate, n_estimators, max_depth.
📉 Overfitting Control
Use max_depth, min_data_in_leaf, feature_fraction, and bagging_fraction to reduce overfitting.
📂 File Format
Model can be saved in text or binary format using booster.save_model() method.
🧮 Categorical Features
Use categorical_feature argument to handle non-numeric features natively without one-hot encoding.
💻 GPU Support
Enable with device = 'gpu' for faster training on large datasets.
🔄 Integration
Integrates with scikit-learn API and supports use with libraries like Optuna and MLflow for hyperparameter tuning and tracking.