hua-zhou / constrainedlasso.jl

Algorithms for fit constrained lasso

ConstrainedLasso

Documentation	Build Status	Code Coverage

ConstrainedLasso.jl implements algorithms for fitting the constrained lasso problem

where is the response vector, is the design matrix of predictor or covariates, is the vector of unknown regression coefficients, and is a tuning parameter that controls the amount of regularization.

Installation

Within Julia, use the package manager to install ConstrainedLasso:

Pkg.clone("git://github.com/Hua-Zhou/ConstrainedLasso.jl.git")

This package supports Julia v0.6.

Documentation

Citation

The original method paper on the constrained lasso is

James, G. M., Paulson, C. and Rusmevichientong, P. (2013). "Penalized and constrained regression," mimeo, Marshall School of Business, University of Southern California. http://www-bcf.usc.edu/~gareth/research/PAC.pdf

If you use ConstrainedLasso package in your research, please cite the following paper on the algorithms:

Gaines, B., Kim, J. and Zhou, H. (2018). “Algorithms for Fitting the Constrained Lasso,” under revision.

constrainedlasso.jl's Issues

Updating to Julia 1.6 ?

Hey,

How hard would it be to update this package to work with Julia 1.6 ? There are no other packages on the current julia plateform that handle constraint lasso so it would definitely find it's usecases (i would use it ^^).

Edit: See this thread https://discourse.julialang.org/t/what-would-it-take-to-bring-this-julia-0-6-package-to-julia-1-6/59652/21
And also my fork https://github.com/lrnv/ConstrainedLasso.jl where we are starting to do updates to your codebase to make it work on 0.7, before moving it directly to 1.6

We still have a few bugs however.

Edit: It seems like the test Test lsq_classopath: sum-to-zero constraint does not work on my version, for a reason i cannot understand the beta vector has a full line of NANs, maiking hte test fail.

Problem Status Error; Problem Status Suboptimal

I am using this package to estimate a lasso that is a polynomial of x1 and x2 but is constrained to be monotone in both x1 and x2. Even for a simple second-order polynomial (y=b_0 + b_1 x_1 + b_2 x_2 + b_3 x_1 x_2 + b_4 x_1^2 + b_5 x_2^2), I consistently get "Problem status Error; solution may be inaccurate" and "Problem status Suboptimal; solution may be inaccurate." Is this intended? I would have expected the algorithms to perform well on such a simple problem. The output shapes of the given solution do appear monotone and appear to have reasonable shape, but I want to know if there is more I can do to ensure that I am finding the best solution. Thank you!