R: Get reference model structure

get-refmodel {projpred}

R Documentation

Get reference model structure

Description

Generic function that can be used to create and fetch the reference model structure for all those objects that have this method. All these implementations are wrappers to the init_refmodel-function so the returned object has the same type.

Usage

get_refmodel(object, ...)

## S3 method for class 'refmodel'
get_refmodel(object, ...)

## S3 method for class 'vsel'
get_refmodel(object, ...)

## Default S3 method:
get_refmodel(
  object,
  data,
  y,
  formula,
  ref_predfun,
  proj_predfun,
  div_minimizer,
  fetch_data,
  family = NULL,
  wobs = NULL,
  folds = NULL,
  cvfits = NULL,
  offset = NULL,
  cvfun = NULL,
  dis = NULL,
  ...
)

## S3 method for class 'stanreg'
get_refmodel(
  object,
  data = NULL,
  ref_predfun = NULL,
  proj_predfun = NULL,
  div_minimizer = NULL,
  folds = NULL,
  ...
)

init_refmodel(
  object,
  data,
  formula,
  family,
  ref_predfun = NULL,
  div_minimizer = NULL,
  proj_predfun = NULL,
  folds = NULL,
  extract_model_data = NULL,
  cvfun = NULL,
  cvfits = NULL,
  dis = NULL,
  ...
)

Arguments

`object`	Object on which the reference model is created. See possible types below.
`...`	Arguments passed to the methods.
`data`	Data on which the reference model was fitted.
`y`	Target response.
`formula`	Reference model's lme4-like formula.
`ref_predfun`	Prediction function for the linear predictor of the reference model.
`proj_predfun`	Prediction function for the linear predictor of the projections.
`div_minimizer`	Maximum likelihood estimator for the underlying projection.
`fetch_data`	Wrapper function for fetching the data without directly accessing it. It should have a prototype fetch_data(data, data_points, newdata = NULL), where data_points is a vector of data indices and newdata, if not NULL, is a data frame with new data for testing.
`family`	A family object that represents the observation model for the reference model.
`wobs`	A weights vector for the observations in the data. The default is a vector of ones.
`folds`	Only used for K-fold variable selection. It is a vector of fold indices for each data point in data.
`cvfits`	Only used for K-fold variable selection. A list of K-fold fitted objects on which reference models are created.
`offset`	A vector of offsets per observation to add to the linear predictor.
`cvfun`	Only used for K-fold variable selection. A function that, given a folds vector, fits a reference model per fold and returns the fitted object.
`dis`	A dispersion vector for each observation.
`extract_model_data`	A function with prototype extract_model_data(object, newdata, wrhs, orhs), where object is a reference model fit, newdata is either NULL or a data frame with new observations, wrhs is a right hand side formula to recover the weights from the data frame and orhs is a right hand side formula to recover the offset from the data frame.

Value

An object of type refmodel (the same type as returned by init_refmodel) that can be passed to all the functions that take the reference fit as the first argument, such as varsel, cv_varsel, project, proj_predict and proj_linpred.

Examples


if (requireNamespace('rstanarm', quietly=TRUE)) {
  ### Usage with stanreg objects
  dat <- data.frame(y = rnorm(100), x = rnorm(100))
  fit <- rstanarm::stan_glm(y ~ x, family = gaussian(), data = dat)
  ref <- get_refmodel(fit)
  print(class(ref))

  # variable selection, use the already constructed reference model
  vs <- varsel(ref) 
  # this will first construct the reference model and then execute 
  # exactly the same way as the previous command (the result is identical)
  vs <- varsel(fit) 
}

[Package projpred version 2.0.2 Index]