Predict from a mixqr fit

Usage

# S3 method for class 'mixqr'
predict(
  object,
  newdata = NULL,
  type = c("quantile", "quantile_byclass", "posterior", "class", "density"),
  ...
)

Arguments

object: A "mixqr" object.
newdata: Optional data frame. If omitted, the training data are used.
type: One of "quantile" (per-component conditional tau-quantiles, an n x m matrix), "quantile_byclass" (the conditional tau-quantile of each point's most likely component, a vector), "posterior" (responsibilities p_ij), "class" (hard labels), or "density" (the component error-density objects).
...: Unused.

Value

A matrix, vector, or list depending on type.

Classifying genuinely new x

"posterior", "class", and "quantile_byclass" require the RESPONSE in newdata, because component membership is inferred from the residual densities. This core therefore cannot assign a class to a covariate vector x alone; covariate-based gating (membership as a function of x) is provided by the location-varying-gating extension (QMM sub-project 03).