predict.svem_model: Predict Method for SVEM Models

In SVEMnet: Self-Validated Ensemble Models with Lasso and Relaxed-Elastic Net Regression

Predict Method for SVEM Models

Description

Generates predictions from a fitted svem_model.

Usage

## S3 method for class 'svem_model'
predict(
  object,
  newdata,
  debias = FALSE,
  se.fit = FALSE,
  agg = c("parms", "mean"),
  ...
)

Arguments

object	An object of class svem_model (created by SVEMnet()).
newdata	A data frame of new predictor values.
debias	Logical; default is FALSE. If TRUE, apply the linear calibration fit (y ~ y_pred) learned during training when available.
se.fit	Logical; if TRUE, returns standard errors (default FALSE).
agg	Aggregation method for ensemble predictions. One of "parms" (default) or "mean". "parms" uses the aggregated coefficients stored in object$parms (or parms_debiased if debias=TRUE). "mean" averages predictions from individual bootstrap members equally and optionally applies the debias calibration.
...	Additional arguments (currently unused).

Details

The function uses the training terms, factor levels (xlevels), and contrasts saved by SVEMnet(). The terms object environment is set to baseenv() to avoid unexpected lookup of objects in the original environment.

Column handling is strict but robust:

• We require that the set of columns produced by model.matrix on newdata matches the set used in training.

• If model.matrix drops some training columns (e.g., a factor collapses to a single level in newdata), we add those missing columns as zeros so coefficients align.

• We then reorder columns to the exact training order before multiplying.

Rows in newdata that contain unseen factor levels will yield NA predictions (and NA standard errors if se.fit=TRUE); a warning is issued indicating how many rows were affected.

When agg="mean", the predictive SE is the bootstrap spread (row-wise sd of member predictions). If debias=TRUE and the calibration slope is available and finite, SEs are scaled by abs(slope).

Value

A numeric vector of predictions, or a list with components fit and se.fit when se.fit = TRUE.

Examples

set.seed(1)
n  <- 40
X1 <- rnorm(n); X2 <- rnorm(n); X3 <- rnorm(n)
y  <- 1 + 0.8*X1 - 0.5*X2 + 0.2*X3 + rnorm(n, 0, 0.3)
dat <- data.frame(y, X1, X2, X3)
fit <- SVEMnet(y ~ (X1 + X2 + X3)^2, dat, nBoot = 30, relaxed = TRUE)
pred <- predict(fit, dat)
head(pred)

# With SEs and debias
out <- predict(fit, dat, debias = TRUE, se.fit = TRUE, agg = "mean")
str(out)

SVEMnet documentation built on Sept. 9, 2025, 5:38 p.m.