Model predictions based on a fitted ncvreg object

Similar to other predict methods, this function returns predictions from a fitted ncvreg object.

Usage

# S3 method for class 'cv.ncvreg'
predict(
  object,
  X,
  type = c("link", "response", "class", "coefficients", "vars", "nvars"),
  which = object$min,
  ...
)

# S3 method for class 'cv.ncvreg'
coef(object, which = object$min, ...)

# S3 method for class 'cv.ncvsurv'
predict(
  object,
  X,
  type = c("link", "response", "survival", "median", "coefficients", "vars", "nvars"),
  which = object$min,
  ...
)

# S3 method for class 'ncvreg'
predict(
  object,
  X,
  type = c("link", "response", "class", "coefficients", "vars", "nvars"),
  lambda,
  which = 1:length(object$lambda),
  ...
)

# S3 method for class 'ncvreg'
coef(object, lambda, which = 1:length(object$lambda), drop = TRUE, ...)

Arguments

object

Fitted ncvreg model object.

X

Matrix of values at which predictions are to be made. Not used for type="coefficients" or for some of the type settings in predict.

type

Type of prediction:

link returns the linear predictors
response gives the fitted values (i.e., with the inverse link transformation applied)
class returns the binomial outcome with the highest probability
coefficients returns the coefficients
vars returns a list containing the indices and names of the nonzero variables at each value of lambda
nvars returns the number of nonzero coefficients at each value of lambda.

which

Indices of the penalty parameter lambda at which predictions are required. By default, all indices are returned. If lambda is specified, this will override which.

...

Not used.

lambda

Values of the regularization parameter lambda at which predictions are requested. For values of lambda not in the sequence of fitted models, linear interpolation is used.

drop

If coefficients for a single value of lambda are to be returned, reduce dimensions to a vector? Setting drop=FALSE returns a 1-column matrix.

Value

The return value depends on type. Unless otherwise noted, predictions are returned as a matrix with one row per observation in X (or per observation in the original dataset if X is not provided) and one column per value of lambda. If lambda has length 1, the result is simplified to a vector with one element per observation.

type = "link": A numeric matrix of linear predictors.
type = "response": A numeric matrix of fitted values.
type = "class": A numeric matrix of 0/1 values indicating the class with the highest probability.
type = "coefficients": A numeric matrix of estimated coefficients. One row per column of X and one column per value of lambda.
type = "vars": A list with an element for each lambda. Each element is a named vector of integers corresponding to the indices with nonzero coefficients.
type = "nvars": A numeric vector with an element for each value of lambda indicating the number of nonzero coefficients.

References

Breheny P and Huang J. (2011) Coordinate descent algorithms for nonconvex penalized regression, with applications to biological feature selection. Annals of Applied Statistics, 5: 232-253. doi:10.1214/10-AOAS388

Examples