Talk abstract: How to become a Kaggle #1: An introduction to model stacking
Ever wondered how Kaggle masters combine hundreds of different machine learning models to win modelling competitions? Ever wondered how to become ranked #1 on Kaggle? StackNet has helped me do that! StackNet is an open-source, scalable and automated meta-modelling framework that combines various supervised models to improve performance.Written in Java, this library automates many of the laborious aspects of building stacking models, so that you can focus on the important parts and move higher up the Kaggle leaderboards. I will explain some of the considerations for running StackNet and show how I have used it to win Kaggle competitions and generate value for dunnhumby.
Bio: Marios Michailidis is Manager of Data Science at Dunnhumby and a former Kaggle #1. During his part-time PhD in machine learning at University College London (UCL), he has focussed on improving supervised modelling using StackNet – A computational, scalable, analytical meta-modelling framework. In his spare time he has also created KazAnova, a GUI for analytics made in Java.