Solution paths for the generalized lasso with applications to spatially varying coefficients regression

Abstract

Penalized regression can improve prediction accuracy and reduce dimension. The generalized lasso problem is used in many applications in various fields. The generalized lasso penalizes a linear transformation of the coefficients rather than the coefficients themselves. The proposed algorithm solves the generalized lasso problem and provides the full solution path. A confidence set can then be constructed on the generalized lasso parameters based on the modified residual bootstrap lasso. The approach is demonstrated using spatially varying coefficients regression, and it is shown to be both accurate and efficient compared to previous work.