Abstract

Algorithm selection and hyperparameter tuning remain two of the mostchallenging tasks in machine learning. The number of machine learningapplications is growing much faster than the number of machine learningexperts, hence we see an increasing demand for efficient automation of learningprocesses. Here, we introduce OBOE, an algorithm for time-constrained modelselection and hyperparameter tuning. Taking advantage of similarity betweendatasets, OBOE finds promising algorithm and hyperparameter configurationsthrough collaborative filtering. Our system explores these models under timeconstraints, so that rapid initializations can be provided to warm-start morefine-grained optimization methods. One novel aspect of our approach is a newheuristic for active learning in time-constrained matrix completion based onoptimal experiment design. Our experiments demonstrate that OBOE deliversstate-of-the-art performance faster than competing approaches on a test bed ofsupervised learning problems.

 

Introduction (beta)

None

 

Conclusion (beta)

None