Ensemble Methods

The most celebrated ensemble in machine learning never went into production. Netflix had offered a million dollars to anyone who could predict its customers' film ratings 10% more accurately, in root-mean-squared error (RMSE), than its own recommender, Cinematch. The prize went to BellKor's Pragmatic Chaos, a merger of three teams, whose entry scored an RMSE of 0.8567 on the hidden test set, 10.06% better than Cinematch; a rival coalition, The Ensemble, had matched that score and lost because it submitted 20 minutes later. The winning entry combined the predictions of many models from its three teams, and did part of that combining with gradient-boosted decision trees.