Large Scale Huber Regression
Dajiang Lei, Zhijie Jiang, Meng Du, Hao Chen, Yu Chuan Wu · 2018
Huber regression is a robust linear regression, which can learn appropriate model on the noisy data scenario. In the era of big data, the training data contains many normal data. In order to solve the Huber regression model in large-scale data, we transform the Huber regression problem into a global consensus problem and a sharing problem. Two different parallel solutions to the Huber regression problem are implemented based on Alternating Direction Method of Multipliers (ADMM) algorithm in Spark distributed computing framework. The experimental results show that the Huber regression can be solved quickly and maintain high precision by choosing appropriate parallelism.