Improving Predictions of Long Sequences by Hyperparameter Tuning

Abstract

The problem of forecasting long sequences is important in many different domains. Proper selection of the hyperparameters when a machine learning approach is applied could make the difference between adequate and inadequate model. Several algorithms for automatic hyperparameters tuning were evaluated and compared with baseline selection. As a result, recommendations have been made. Some of the intuitive assumptions for the baseline model proved to be wrong.