Andrej Tschalzev

Searching for a great alternative to Gradient-Boosted Decision Trees such as #XGBoost & #CatBoost? 🌳 Unveiling GRANDE: a novel gradient-based decision tree ensemble for tabular data!  
🚀 Ready to dive in? pip install GRANDE
(1/3) 
github.com/s-marton/GRANDE

@ChristophMolnar Possible reasons: Missing theoretical foundation, significance tests,... Lack of ML knowledge of those seeking for true insights, while ML people hunt for performance metrics Flexible models find different solutions to predict a target, while we seek robust insights

@ChristophMolnar Great idea! In social science most publications use linear and logistic regression to quantify relationships between variables. The reason is the great theoretical foundation. There is huge potential for nonlinear ML methods. The question is why these methods are so rarely used?

I'm brainstorming my 3rd book right now. Rough idea: How to Turn Predictions Into Insights Think supervised ML plus tools and constraints needed to gain insights. Audience: (data) scientists who use machine learning as means to answer (scientific) questions. 1/n