SSGL: Estimate grouped spike and slab lasso regression model
In jantonelli111/SSGL: Spike and Slab Group Lasso

Description Usage Arguments Value Examples

View source: R/Main.R

This function takes in an outcome and covariates, and estimates the posterior mode of the spike and slab group lasso penalty

SSGL(Y, X, lambda1, lambda0, groups, a = 1, b = length(unique(groups)),
  updateSigma = TRUE, M = 10, error = 0.001, betaStart = rep(0,
  dim(X)[2]), sigmasqStart, theta, printWarnings = TRUE,
  forceGroups = c())

`Y`	The outcome to be analyzed
`X`	An n by p matrix of covariates
`lambda1`	Prior parameter for the slab component of the prior
`lambda0`	Prior parameter for the spike component of the prior
`groups`	A vector of length p denoting which group each covariate is in
`a`	First hyperparameter for the beta prior denoting the prior probability of being in the slab
`b`	Second hyperparameter for the beta prior denoting the prior probability of being in the slab
`M`	Positive number less than p indicating how often to update theta and sigma. There is no need to change this unless trying to optimize computation time
`betaStart`	Starting values for beta vector. There is no need to change this unless for some specific reason
`theta`	Value of the sparsity parameter theta. There is no need to select a value for this parameter unless the sparsity is known a priori. If left blank, the function will update theta automatically
`printWarnings`	Print warning messages about whether the optimization has converged
`forceGroups`	A vector containing the indices of any groups you wish to automatically include in the model and not penalize
`UpdateSigma`	True or False indicating whether to estimate the residual variance
`sigmaStart`	Starting value for sigma. There is no need to change this unless for some specific reason

A list of values containing the regression coefficients, the intercept, the estimate of the residual variance (this is simply the marginal variance of Y if UpdateSigma=FALSE), An estimate of theta the prior probability of entering into the slab, and the number of iterations it took to converge

## Here we generate 200 samples from 100 covariates
n = 200
G = 100
x = mvtnorm::rmvnorm(n, sigma=diag(G))


X = matrix(NA, nrow=n, ncol=G*2)

for (g in 1 : G) {
  X[,2*(g-1) + 1] = x[,g]
  X[,2*g] = x[,g]^2
}

Y = 200 + x[,1] + x[,2] + 0.6*x[,2]^2 + rnorm(n, sd=1)

## Now fit model for chosen lambda0 and lambda1 values
modSSGL = SSGL(Y=Y, X=X, lambda1=.1, lambda0=10, 
groups = rep(1:G, each=2))

modSSGL