projpred: proj-pred – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

proj-pred

Extract draws of the linear predictor and draw from the predictive distribution of the projected submodel

Description

proj_linpred extracts draws of the linear predictor and proj_predict draws from the predictive distribution of the projected submodel or submodels. If the projection has not been performed, the functions also perform the projection.

Usage

proj_linpred(
  object,
  newdata,
  offsetnew = NULL,
  weightsnew = NULL,
  nterms = NULL,
  transform = FALSE,
  integrated = FALSE,
  seed = NULL,
  ...
)

proj_predict(
  object,
  newdata,
  offsetnew = NULL,
  weightsnew = NULL,
  nterms = NULL,
  ndraws = 1000,
  seed = NULL,
  ...
)

Arguments

`object`	Either an object returned by varsel, cv_varsel or init_refmodel, or alternatively any object that can be converted to a reference model.
`newdata`	The predictor values used in the prediction. If `solution_terms` is specified, then `newdata` should either be a dataframe containing column names that correspond to `solution_terms` or a matrix with the number and order of columns corresponding to `solution_terms`. If `solution_terms` is unspecified, then `newdata` must either be a dataframe containing all the column names as in the original data or a matrix with the same columns at the same positions as in the original data.
`offsetnew`	Offsets for the new observations. By default a vector of zeros. By default we take the weights from newdata as in the original model. Either NULL or right hand side formula.
`weightsnew`	Weights for the new observations. For binomial model, corresponds to the number trials per observation. For `proj_linpred`, this argument matters only if `newdata` is specified. By default we take the weights from newdata as in the original model. Either NULL or right hand side formula.
`nterms`	Number of terms in the submodel (the variable combination is taken from the variable selection information). If a vector with several values, then results for all specified model sizes are returned. Ignored if `solution_terms` is specified. By default use the automatically suggested model size.
`transform`	Should the linear predictor be transformed using the inverse-link function? Default is `FALSE`. For `proj_linpred` only.
`integrated`	If `TRUE`, the output is averaged over the parameters. Default is `FALSE`. For `proj_linpred` only.
`seed`	An optional seed to use for drawing from the projection. For `proj_predict` only.
`...`	Additional argument passed to project if `object` is an object returned by varsel or cv_varsel.
`ndraws`	Number of draws to return from the predictive distribution of the projection. The default is 1000. For `proj_predict` only.

Value

If the prediction is done for one submodel only (nterms has length one or solution_terms is specified) and newdata is unspecified, a matrix or vector of predictions (depending on the value of integrated). If newdata is specified, returns a list with elements pred (predictions) and lpd (log predictive densities). If the predictions are done for several submodel sizes, returns a list with one element for each submodel.

Examples

if (requireNamespace('rstanarm', quietly=TRUE)) {
  ### Usage with stanreg objects
  n <- 30
  d <- 5
  x <- matrix(rnorm(n*d), nrow=n)
  y <- x[,1] + 0.5*rnorm(n)
  data <- data.frame(x,y)
  
  fit <- rstanarm::stan_glm(y ~ X1 + X2 + X3 + X4 + X5, gaussian(), data=data, chains=2, iter=500)
  vs <- varsel(fit)
  
  # compute predictions with 4 variables at the training points
  pred <- proj_linpred(vs, newdata = data, nv = 4)
  pred <- proj_predict(vs, newdata = data, nv = 4)
}

projpred

Projection Predictive Feature Selection

v2.0.2

GPL-3

Authors

Juho Piironen [aut], Markus Paasiniemi [aut], Alejandro Catalina [cre, aut], Aki Vehtari [aut], Jonah Gabry [ctb], Marco Colombo [ctb], Paul-Christian Bürkner [ctb]

Initial release