Seminar 12 -- Winners and Winning Solutions from the ASHRAE Great Energy Predictor III Machine Learning Competition PDF

This product is a zip file that contains files that consist of PowerPoint slides synchronized with the audio-recording of the speaker, PDF files of the slides, and audio only (mp3 format) as noted.

This seminar provides a formal recognition of the models developed in the ASHRAE-hosted Great Energy Predictor III machine learning competition by giving members from the top three winning teams the opportunity to show their solutions. The winners will discuss the preprocessing, feature engineering, model selection and structure, and post-processing techniques that they used in the competition. In addition, a large overview of results was undertaken by the competition technical team that illustrates the accuracy/complexity balance of the solutions as well as the online resources generated by the competition.

• 1. Overview Analysis of the Great Energy Predictor (GEP) III Competition Models
  Clayton Miller, Ph.D., Member, National University of Singapore, Singapore, Singapore
• 2. GEP III First-Place Solution - Group-Based Ensembles and Strategic Pre-Processing
  Matthew Motoki, Iterable, San Francisco, CA
• 3. GEP III Second-Place Solution - Intensive Pre-Processing and Huge Xgboost, Lightgbm, Catboost, and Ffnn Ensemble
  Rohan Rao, DSc, H2O.ai, Bangalore, India
• 4. GEP III Third-Place Solution - Xgboost and Lightgbm with Weighted Post-Processing
  Xavier Capdepon, Xavier Capdepon, New York, NY