FLASH: Fast Bayesian Optimization for Data Analytic Pipelines
Yuyu Zhang*, Georgia Institute of Technolog; Mohammad Bahadori, Georgia Institute of Technology; Hang Su, Georgia Institute of Technology; Jimeng Sun, Georgia Institute of Technology
Modern data science relies on data analytic pipelines to organize interdependent computational steps. Such analytic pipelines often involve diﬀerent algorithms across multiple steps, each with its own hyperparameters. To achieve the best performance, it is often critical to select optimal algorithms and to set appropriate hyperparameters, which requires large computational eﬀorts. Bayesian optimization provides a principled way for searching optimal hyperparameters for a single algorithm. However, many challenges remain in solving pipeline optimization problems with high-dimensional and highly conditional search space. In this work, we propose Fast LineAr SearcH (FLASH), an eﬃcient method for tuning analytic pipelines. FLASH is a two-layer Bayesian optimization framework, which ﬁrstly uses a parametric model to select promising algorithms, then computes a nonparametric model to ﬁne-tune hyperparameters of the promising algorithms. FLASH also includes an eﬀective caching algorithm which can further accelerate the search process. Extensive experiments on a number of benchmark datasets have demonstrated that FLASH signiﬁcantly outperforms previous state-of-the-art methods in both search speed and accuracy. Using 50% of the time budget, FLASH achieves up to 20% improvement on test error rate compared to the baselines. FLASH also yields state-of-the-art performance on a real-world application for healthcare predictive modeling.
Filed under: Big Data