Constructor for inferrers — as_inferrer • selectInferToolkit

Create an inferrer wrapper around post-selection inference algorithms so they can be used with the selectInferToolkit package and can harmonize results across various selection algorithms (inferrers). This is not a user-facing function.

This function performs a selector's default inference method. Alternative inference methods may be available via infer_*, which you should use instead if you want to do specific tuning.

A wrapper for the selectiveInference functions on selector objects

UPSI is sometimes referred to as "hybrid-OLS", but essentially we re-fit the selected model to the data as though we never used that same data to fit the model. It is common, easy, and ill-advised.

Usage

as_inferrer(
  x,
  name,
  label = name,
  nonselection,
  inferences,
  conf.level,
  selector,
  meta = list()
)

infer(
  object,
  data,
  nonselection = c("ignored", "uncertain_nulls", "confident_nulls"),
  ...
)

infer_pipe(object, data, conf.level = 0.95, ...)

infer_selective(
  object,
  data,
  nonselection = c("ignored", "confident_nulls", "uncertain_nulls"),
  conf.level = 0.95,
  use_cv_sigma = FALSE,
  ...
)

infer_upsi(
  object,
  data,
  nonselection = c("ignored", "confident_nulls", "uncertain_nulls"),
  conf.level = 0.95
)

Arguments

x: a slot for the main inferential object
name: brief name for method
label: label for method
nonselection: A character string specifying how to handle variables not selected by model selection procedure. One of "ignored", "confident_nulls" or "uncertain_nulls" supported
inferences: a slot for post-selection inferences that should have confidence intervals (at least)
conf.level: .95 by default
selector: a carried-forward selector object
meta: a list containing important meta-information from the method
object: a selector object
data: data must be passed to infer methods
...: arguments passed to selectiveInference function(s)
use_cv_sigma: estimate Sigma via CV (if FALSE, uses SI defaults)

Value

An S3 object (list) of class inferrer

A list of class infer_* containing:'

ci_avg_ratio: Average CI length across all variables in model
ci_median_ratio: median CI length across all variables in model
nonselection: method chosen to deal with non selection
infmethod: Inference method chosen
selection_method: Stepwsie,returned for selector_stepwise_ic class only
direction: the mode of step wise search, returned for selector_stepwise_ic class only
penalty: penalty used (AIC or BIC), returned for selector_stepwise_ic class only
lambda: selected lambda for inference , either "lambda.min" or "lambda.1se"; returned for selector_pen class only
alpha: selected alpha for inference, returned for selector_pen class only
B: The number of bootstrap replicates used (only for bootstrap selection method)

an inferror object

inferrer object