R/setup_parameter.R

In r08in/RobustCD: Penalized Adaptive Weighted Least Squares Regression

setup_parameter = function(x, y, nlambda1, nlambda2, lambda1.min=0.05,
                           lambda2.min=1e-03, penalty.factor1, penalty.factor2) {
    n = length(y)
    p =dim(x)[2]
    lambda2Max = max(y^2 /penalty.factor2/n)
    lambda1Max = max(abs(t(x) %*% y/n) / penalty.factor1)  # max |betaj|*|xj'y/n|
    lambda2 = logSeq2(lambda2Max, lambda2Max * lambda2.min, nlambda2)
    lambda1 = logSeq2(lambda1Max, lambda1Max * lambda1.min, nlambda1)
    return(list(lambda2 = lambda2, lambda1 = lambda1))
}

logSeq2 = function(smax, smin, n) {
  if(smin==0){
    c(exp(seq(log(smax), log(1e-10), length.out=n-1)),0)
  }else{
    exp(seq(log(smax), log(smin), length.out=n))
  }
  
}