MWLasso-package: A penalized Moving-Window Lasso (MWLasso) method for...

In mllbao/MWLasso: A penalized Moving-Window Lasso (MWLasso) method for genome-wide association studies.

Description

Find coefficients for a regression model with moving-window Lasso penalty (MWLasso). The model is marginalized so that missing values can be accommendated.

Usage

MWLasso(X, Y, lambda, eta, d, method='linear', epson = 1e-10, M = 100)
MW_parameters(X, Y, d, number, gamma2, method="linear", epson = 1e-10, M = 100)

Arguments

X	The design matrix which can include missing values.
Y	The response variable.
lambda	The L1 penalty tuning parameter.
eta	The moving-window penalty tuning parameter.
d	The size of the moving-window penalty.
method	The regression method. Method can be linear or logistic.
number	The pre-determined number of non-zero coefficients.
gamma2	The proportion of the tuning parameter for moving-window part and L1 penalty part. gamma2 is the ratio of lambda/(lambda+eta).
epson	Convergence criterion. The iteration will stop if the relative change is smaller than epson.
M	The maximum number of iterations.

Details

The function minimizes 1/(2n)*MLS + lambda*L1 + eta/(2(d-1))*MW. Here MLS is the marginalized least squares, L1 is the L1 penalty in Lasso, and MW is the moving-window penalty.

Value

MWLasso returns:

beta	The coefficients estimates.

MW_parameters returns:

lambda	The tuning parameter for L1 penalty.
eta	The tuning parameter for moving-window penalty.

Thus under such lambda and eta, MWLasso will return beta which satisfies the pre-determined number of non-zero coefficients.

Author(s)

Minli Bao <[e-mail]>

References

Genome-wide association studies using a penalized moving-window regression. Minli Bao and Kai Wang. Submitted for review.

Examples

n = 100
p = 200
X = matrix(rnorm(n*p, mean = 0, sd = 1), nrow = n, ncol = p)
X[2,1] = NA; X[80:90,1]= NA; 
Y =  rnorm(n, mean = 0, sd = 1)

number = 20
gamma2 = 0.1
d = 4
params = MW_parameters(X, Y, d, number, gamma2, method="linear", epson = 1e-10, M = 100)
lambda = params[1]
eta = params[2]
beta.hat = MWLasso(X, Y, lambda, eta, d, method='linear')

mllbao/MWLasso documentation built on May 23, 2019, 2:12 a.m.