update_each_effect_rss.R

# @title Update each effect once.
# @param R a p by p symmetric and positive semidefinite correlation matrix.
# @param z a p vector
# @param s_init a list with elements sigma2, V, alpha, mu, Xr
# @param Sigma sigma2*R + lambda I
# @param estimate_prior_variance boolean indicating whether to estimate
#   prior variance
# @param check_null_threshold float a threshold on the log scale to
#   compare likelihood between current estimate and zero the null
# 
#' @importFrom Matrix diag
update_each_effect_rss = function (R, z, s_init, Sigma,
                                   estimate_prior_variance = FALSE,
                                   estimate_prior_method = "optim",
                                   check_null_threshold = 0) {

  if (!estimate_prior_variance)
    estimate_prior_method = "none"
  
  # Repeat for each effect to update.
  s = s_init
  L = nrow(s$alpha)
  if(L > 0)
    for (l in 1:L) {
        
      # Remove lth effect from fitted values.
      s$Rz = s$Rz - R %*% (s$alpha[l,] * s$mu[l,])

      #compute residuals
      r = z - s$Rz
      res = single_effect_regression_rss(as.vector(r),Sigma,s$V[l],s$pi,
              estimate_prior_method,check_null_threshold)
      
      # Update the variational estimate of the posterior mean.
      s$mu[l,]    = res$mu
      s$alpha[l,] = res$alpha
      s$mu2[l,]   = res$mu2
      s$V[l]      = res$V
      s$lbf[l]    = res$lbf_model
      s$KL[l]     = -res$lbf_model +
        SER_posterior_e_loglik_rss(R,Sigma,r,res$alpha * res$mu,
                                   res$alpha * res$mu2)
      s$Rz = s$Rz + R %*% (s$alpha[l,] * s$mu[l,])
    }
  s$Rz = unname(as.matrix(s$Rz))
  return(s)
}