Run multiple csmoothEM initializations in parallel and summarize results

Runs initialize_csmoothEM with multiple initialization methods, each for num_iter warm-start iterations, and returns a named list of fits. A per-method summary table is attached as an attribute "summary".

The summary table reports the last values of elbo_trace, loglik_trace, and ml_trace. In the current codebase, ml_trace is the collapsed objective \(\mathcal{C}\) (Laplace-exact in the Gaussian csmooth setting).

parallel_initial_csmoothEM(
  X,
  methods = c("PCA", "tSNE", "random", "fiedler", "pcurve"),
  num_iter = 1,
  num_cores = 2,
  K = NULL,
  adaptive = "prior",
  lambda_min = 1e-08,
  lambda_max = 1e+08,
  sigma_update = c("mstep", "ml"),
  sigma_min = 1e-10,
  sigma_max = 1e+10,
  seed = NULL,
  quiet = TRUE,
  ...
)

Arguments

X

Numeric matrix (n x d).

methods

Character vector of initialization methods. Default includes "PCA", "tSNE", "random", "fiedler", "pcurve", "isomap". See initialize_csmoothEM for details.

num_iter

Integer \(\ge 1\). Number of warm-start iterations to run inside initialize_csmoothEM for each method.

num_cores

Integer \(\ge 1\). Number of cores to use. On non-Windows systems, uses parallel::mclapply; on Windows, falls back to a PSOCK cluster.

K

Optional integer \(\ge 2\). Number of mixture components. If NULL, the default logic in initialize_csmoothEM is used.

adaptive

Character. One of "none", "prior", "ml". Logical values are accepted for backward compatibility: TRUE is treated as "prior" and FALSE as "none".

lambda_min, lambda_max

Positive bounds for lambda_vec (passed to initialize_csmoothEM).

sigma_update

Character. Only used when adaptive="ml" (passed to initialize_csmoothEM). Either "mstep" or "ml"; see do_csmoothEM_ml_collapsed.

sigma_min, sigma_max

Positive bounds for sigma2 when adaptive="ml" and sigma_update="ml".

seed

Optional integer seed. If provided, a different derived seed is used per method.

quiet

Logical; reserved for future use.

...

Additional arguments passed to initialize_csmoothEM (and downstream ordering routines).

Value

A named list of fits, with names equal to methods. Each entry is either a csmooth_em object (on success) or NULL (on failure). The list has an attribute "summary" which is a data.frame with one row per method:

  • method: method name

  • success: logical

  • n,d,K: scalar integers from summary(csmooth_em) when available

  • elbo_last: last penalized ELBO (or NA)

  • obj_last: last penalized observed objective (or NA)

  • ml_last: last collapsed objective \(\mathcal{C}\) (or NA)

  • error: error message if failed