Progressive-resolution initialization for SmoothEM via kriging on a final grid

Coarse-to-fine continuation scheme for initializing SmoothEM means over nested grids.

progressive_smoothEM(
  data,
  m_max = 6,
  lambda_final = 500,
  q = 2,
  ridge = 0,
  nugget_kriging = 0,
  tol = 0.001,
  max_iter = 1000,
  relative_lambda = TRUE,
  modelName = "EEI",
  coords_show = c(1, 2, 3),
  plot_each_stage = TRUE,
  verbose = TRUE,
  include.data = TRUE
)

Arguments

data

Numeric matrix n x d.

m_max

Finest grid exponent; final grid size is 2^m_max + 1.

lambda_final

Penalty strength on final grid.

q

RW order (e.g. 2 for RW2).

ridge

Ridge added in RW precision construction.

nugget_kriging

Nugget added to Q_UU during kriging solve.

tol

Convergence tolerance on ELBO changes.

max_iter

Maximum number of EM iterations.

relative_lambda

Logical; if TRUE, rescales Q_prior by current marginal variances.

modelName

Covariance model: one of "VVV", "VII", "EII", "EEI".

coords_show

Coords to visualize when plot_each_stage=TRUE.

plot_each_stage

If TRUE, plot kriged mean curves each stage.

verbose

If TRUE, print stage summaries and forward verbose to EM_algorithm().

include.data

If TRUE, include data in returned fits.

Value

List(grid, Q_final_1d, fits, mu_full_history, mu_full_list_final, meta_history, mclust).