Predicted response curves for new covariates — predict.mfsusie • mfsusieR

Projects the posterior coefficient curves through new covariate values to produce predicted response curves on the original Y scale, per outcome. The wavelet pipeline that built the fit (column scaling, padding, DWT) is inverted to yield curves on each outcome's original position grid pos[[m]].

Usage

# S3 method for class 'mfsusie'
predict(object, newx = NULL, ...)

Arguments

object: an mfsusie fit returned by mfsusie().
newx: numeric matrix n_new x p of new covariates on the same scale as the training X. NULL (default) returns the training fitted values (equivalent to fitted(object)).
...: ignored.

Value

list of length M; each element a numeric matrix n_new x T_m of predicted curves on the original position grid for that outcome.

Details

Prediction uses the per-variable alpha-weighted aggregate coefficient b[j, t] = sum_l alpha[l, j] * mu[l, j, t] / csd_X[j], the same form as susieR::predict.susie().