CatBoost - The New Generation of Gradient Boosting

Explore the cutting-edge CatBoost gradient boosting library in this EuroPython 2018 conference talk. Dive into the power of gradient boosting for machine learning tasks with heterogeneous features, noisy data, and complex dependencies. Learn how CatBoost outperforms existing implementations in terms of quality, incorporating categorical features without preprocessing. Discover its advantages, including 20-60 times faster inference, GPU and multi-GPU training capabilities, and scalability across hundreds of machines. Gain insights into the proprietary algorithm behind CatBoost's quality boost, and understand its applications in web search, recommendation systems, and weather forecasting. Compare CatBoost with other gradient boosting libraries, explore its modes for classification, regression, and ranking, and learn about features like SHAP values and the CatBoost Viewer.

Intro
Gradient Boosting
Applications
Neural networks
Algorithm comparison
Symmetric trees
Numerical features
Categorical features support
Classical boosting
Ordered boosting
Modes
Classification
Regression
Ranking
GPU: Comparison with other libraries
Prediction time
SHAP values
CatBoost Viewer
Cross-validation
Reading
Algorithm parameters