Append one coordinate to a csmooth_em fit using a Gamma-based 1D initialization (internal)

Adds a new feature column to a csmooth_em object, initializing the new coordinate's \(\mu_{j\cdot}\) and \(\sigma_j^2\) using score_feature_given_Gamma under the fit's current responsibilities fit$gamma. Then appends the coordinate and sets its lambda to 1 (it can be updated later by do_csmoothEM if adaptive is on).

This is a warm-start operation: it does not reinitialize the whole model.

append_coord_to_fit_csmooth(
  fit,
  xj,
  score_mode = c("ml", "none"),
  rw_q = 2,
  relative_lambda = TRUE,
  lambda_min = 1e-10,
  lambda_max = 1e+10
)

Arguments

fit

A csmooth_em object (homoskedastic).

xj

Numeric vector length n (the new feature column).

score_mode

One of "ml" or "none"; controls how the 1D init is computed.

rw_q

Integer RW rank deficiency along K.

relative_lambda

Logical.

lambda_min, lambda_max

Bounds for lambda optimization when score_mode="ml".

Value

Updated csmooth_em.