initialize_tree_lcm: Initialize the variational inferential algorithm for latent...

In zhenkewu/lotR: Latent class analysis with Observed Trees in R (lotR)

Initialize the variational inferential algorithm for latent class model with tree structured shrinkage

Description

Initialize the variational inferential algorithm for latent class model with tree structured shrinkage

Usage

initialize_tree_lcm(
  Y,
  A,
  Z_obs,
  leaf_ids_units,
  leaf_ids_nodes,
  ancestors,
  h_pau,
  levels,
  vi_params,
  hyperparams,
  hyper_fixed,
  random_init,
  random_init_vals,
  subject_id_list,
  v_units,
  shared_tau
)

Arguments

Y	matrix of binary observations for LCM - rows ordered by leaves in the tree
A	a p by p binary matrix: each row is the ancestors including the node itself; ordered by leaves in the tree
Z_obs	a two-column matrix of integers, first column is subject ids, second column is a mix of NA and integers, NA means unknown class indicators, an integer indicates the known class.
leaf_ids_units	The subject ids in each leaf node (a list)
leaf_ids_nodes	the leaf descendants for each internal or leaf nodes (a list)
ancestors	a numeric vector of ancestor nodes for each leaf node (a list of length equal to the number of leaves)
h_pau	a numeric vector (length = # nodes); the edge length between a node and its parent node. The root node has no parent, because we suggest a separate prior variance for the root node's gamma and alpha, we set the "edge-length" toward root node as 1.
levels	a numeric vector of integers from 1 to L, indicating for each node (leaf or internal node) which set of hyperparameters to use. For example, if we want the root node to have a separate tau_1 and tau_2, can specify it to its own level. Another example would be to have distinct sets of hyperparameters for leaf and non-leaf nodes. The levels are pre-specified, not estimated.
vi_params	the list of variational parameters. mu_gamma, mu_alpha, prob (for s_u), a_t, b_t,sigma_gamma, Sigma_alpha prob,prob_gamma=c(1,rep(0,p-1)) (this is not updated).
hyperparams	the list of hyperparameters, tau_1 and tau_2 - these are initial specifications of the hyperparameters - they are updated by tau_1_t, tau_2_t; psi, g_psi, phi, g_phi (these are not hyperparameters, but are updated in the same function as tau_1_t and tau_2_t), but more frequently.
hyper_fixed	a list of fixed hyperparameters, such as those in the Beta priors for rho, e.g., list(a=c(1,1,99),b=c(1,1,1)), K, number of classes. for three levels, where the first and second levels are uniform Beta, the third level has Beta(99,1) prior - this nearly ensures setting s_u=1.
random_init	logical; TRUE for adding additional variability to the initial values. This is needed if the algorithm needs multiple random starts to choose the best converged values.
random_init_vals	NB: fill out specific elements
subject_id_list, v_units, shared_tau	see update_vi_params()

Value

a list vi_params,hyperparams containing the initial values.

zhenkewu/lotR documentation built on April 24, 2022, 2:36 a.m.