pysmme 1.0

Soft maximin estimation in Python

Efficient C++ procedure for solving the Lasso or SCAD penalized soft maximin problem.

This is a C++ implementation of two proximal gradient based algorithms (NPG and FISTA) that solve different forms of the soft maximin problem from Lund et al., 2022 see https://doi.org/10.1111/sjos.12580. 1) For general group specific design the soft maximin problem is solved using the NPG algorithm. 2) For fixed identical design across groups, the soft maximin problem is solved using either the FISTA algorithm or the NPG algorithm in the following two cases: i) For a tensor structured design matrix the algorithms use array arithmetic to avoid the design matrix and speed computations ii) For a wavelet based design matrix the algorithms use the pyramid algorithm to avoid the design matrix and speed up computations. Multi-threading is possible when openMP is available.