PyData Berlin

Gradient boosting in practice: a deep dive into xgboost

Jaroslaw Szymczak

From variety of classification and regression methods, gradient boosting, and in particular its variation in xgboost implementation, is one of the most convenient to use. Out of the box you can use it as easily as random forest. Due to its nature, when used with decision trees, you don’t need to worry about co-linearities or missing values. No more worrying about normalization, standardization nor any other monotonic transformations on your data. Overfitting prevention with watchlists. Written efficiently in C++ with Python and R bindings and scikit-learn like interface. In this talk we will go deep into how and why xgboost works, why it is present in so many winning Kaggle solutions, what is the meaning of its parameters, how to tune them and how to use it in practice.

About the Speaker: Jaroslaw is a Machine Learning Scientist in OLX Tech Hub Berlin. He has background in analytics and predictive models creation for finance institutions, FMCG and Telecom companies. Currently he is specializing in applying machine learning to detection of unwanted content on OLX classifieds sites across the globe.