Skip to contents

Coefficients and Prediction for Penalized MARS Models

Usage

# S3 method for class 'mars_lasso'
coef(object, s = NULL, standardized = FALSE, ...)

# S3 method for class 'mars_lasso'
predict(
  object,
  newdata = NULL,
  s = NULL,
  type = c("response", "working", "matrix"),
  matrix_action = c("warn", "error", "repair", "none"),
  ...
)

Arguments

object

Fitted mars_lasso object.

s

NULL for the selected penalty, or an exact fitted lambda value. No interpolation or refitting is performed.

standardized

Return coefficients on the optimization scale instead of the original supplied design scale.

...

Unsupported arguments are rejected.

newdata

Raw predictor records. NULL uses the training data. Correlation matrix prediction needs the internal study and effect columns stored on object.

type

response returns effect-scale population predictions; working returns the fitting scale; matrix returns complete named predicted correlation matrices.

matrix_action

warn, error, repair or none, as in predict.mars_rf(). Explicit repair records matrix changes; bounded pairs do not ensure joint validity.

Value

Coefficients as a named numeric vector, or predictions as a vector/list of matrices. Predictions omit BLUPs and do not provide post-selection intervals. A zero coefficient is not a hypothesis-test result. Frozen training transforms and conditional EM moments are used for new records.