Choose the best SmoothEM initialization by ELBO — optimize

Runs parallel_initial() and returns the smooth_em object with the largest last-iteration ELBO.

If plot=TRUE, plots all ELBO traces (different colors) and optionally overlays the penalized objective traces in a second panel.

optimize_initial(
  X,
  methods = c("PCA", "tSNE", "random", "fiedler", "multi_scale", "pcurve"),
  num_iter = 1,
  num_cores = 2,
  m_max = 6,
  K = NULL,
  adaptive = TRUE,
  lambda_min = 1e-08,
  lambda_max = 1e+08,
  plot = FALSE,
  two_panel = FALSE,
  seed = NULL,
  quiet = TRUE,
  ...
)

Arguments

X: Numeric matrix (n x d).
methods: Methods to try (passed to parallel_initial()).
num_iter: Total number of iterations for each fit.
num_cores: Number of cores.
m_max: Used for multi_scale and default K for others.
K: Optional K for non-multi-scale methods.
adaptive: Logical; passed to parallel_initial() / initialize_smoothEM().
lambda_min, lambda_max: Positive bounds for lambda when adaptive=TRUE.
plot: Logical; if TRUE, plot traces.
two_panel: Logical; if TRUE, show ELBO and objective in 2 panels.
seed: Optional base seed.
quiet: Logical.
...: Passed to initialize_smoothEM() via parallel_initial().

Value

A smooth_em object (best by last ELBO). The returned object gains:

$meta$initial_search: list with the full fits, summary table, and options.