adg_em: Estimate Q, screening latent attributes (Alternating Gibbs)

In zhenkewu/slamR: Structured Latent Attribute Models in R

Description

This function implements the Alternating Direction Gibbs EM (ADG-EM) algorithm in the scenario of responses observed over many taxonomies (trees)

Usage

adg_em(
  X,
  Z_ini,
  Q_ini,
  max_iter,
  err_prob,
  must_maxiter = 0,
  D_mat = NULL,
  X1 = NULL,
  model = "DINA"
)

Arguments

X	N by J2 binary data matrix - level 2
Z_ini	N by K initial latent attributes
Q_ini	J by K inital Q matrix
max_iter	maximum iterations (e.g., 50)
err_prob	noise level
must_maxiter	1 to force maxiter; default is 0
D_mat	J1 by J2 binary matrix to indicate children in two-level trees. D_mat is the J1 * J2 binary adjacency matrix specifying how the trees are grown in the second layer, i.e., which second-level responses are "children" of each first-level response. Default is NULL
X1	N by J1 binary data matrix - level 1; default is NULL
model	"DINA" (default) or "DINO"

Value

• Z_est Estimated latent attributes for all people

• Z_candi candidate latent attribute patterns (unique)

• Q_arr a list of Q matrices obtained from the algorithm

• c J dimensional 1-slipping parameter

• g J dimensional guessing parameter

zhenkewu/slamR documentation built on March 8, 2020, 1:31 a.m.