Large-scale distributed solution of packet Dantzig Selector

Large-scale distributed solution of packet Dantzig Selector

Liang of China University of Science and Technology

The main work of this paper is as follows: (1) by using the characteristic of the Dantzig selector to segment linearity, the improved Dasso algorithm is used to solve the Dantzig selector, and the superiority of the improved algorithm is highlighted by the comparison with the linear alternating direction multiplier method. (2) Overcome the difficulty of solving the constraint condition in the packet Dantzig selector, introduce the intermediate variable to simplify, and apply the alternating direction multiplier method (ADMM) and the linear alternating direction multiplier method (LADMM) algorithm, so that the solution of the group Dantzig selector becomes possible. (3) Build the distributed computing platform on the server, create the virtual machine, use the spark to realize the ADMM algorithm of the grouping Dantzig selector, and compare the centralized calculation with the distributed computational efficiency. The group Dantzig selector has been applied to the linear regression model with packet sparsity in feature selection, model prediction and other problems, and the famous examples have brain Wave medicine experiment. First, the brain waves of the human head were measured at 256Hz by the 64 Microelectrode set on the scalp of the human body, and the disease symptom of the sample population was recorded, and the regression model was established. Then the model parameters are calculated according to these data, which can be used to predict the future operation. Distributed computing has important significance to deal with the increasingly large-scale data, which is also the important starting point of this paper.

Large-scale distributed solution of packet Dantzig Selector