init.EM.lasso: Initialization of the shifts using Lasso.

Description

init.EM.lasso does the following regression : ||Y_data-T.delta||_(Sigma_YY^(-1)) + lambda |delta|_1 using the function glmnet::glmnet of package glmnet, through function lasso_regression_K_fixed. T is the incidence matrix of the tree, and delta the vectorial representation of the shifts (see functions incidence.matrix and shifts.list_to_vector for further details).

Usage

init.EM.lasso(
  phylo,
  Y_data,
  Y_data_imp = Y_data,
  Y_data_vec_known = as.vector(Y_data),
  process,
  times_shared = compute_times_ca(phylo),
  distances_phylo,
  nbr_of_shifts,
  K_lag_init = 0,
  use_sigma = TRUE,
  params_sigma = NULL,
  variance.init = diag(1, p, p),
  random.init = FALSE,
  value.root.init = rep(0, p),
  exp.root.init = rep(1, p),
  var.root.init = diag(1, p, p),
  edges.init = NULL,
  values.init = matrix(0, p, length(edges.init)),
  relativeTimes.init = NULL,
  selection.strength.init = 1,
  optimal.value.init = rep(0, p),
  T_tree = incidence.matrix(phylo),
  subtree.list = NULL,
  miss = FALSE,
  sBM_variance = FALSE,
  stationary.root.init = FALSE,
  masque_data,
  independent = FALSE,
  ...
)

Value

params_init the list of initial parameters to be used, in the right format.

Arguments

Y_data: data at the tips.
times_shared: (matrix) : times of shared ancestry, result of function compute_times_ca.
distances_phylo: (matrix) : phylogenetic distance, result of function compute_dist_phy
nbr_of_shifts: number of shifts used in the EM algorithm

Details

A Cholesky decomposition of function Sigma_YY^(-1) is used. lambda is chosen so that delta has the right number of non zero components.