calibration: calibration

In AndrewLawrence/dCVnet: doubly cross-validated elastic-net regularised generalised linear models

calibration

Description

calculates 'weak' (i.e. intercept + slope) calibration for binomial family outcome. This is equivalent to the values returned by the val.prob function in the rms package that accompanies Frank Harrell's Regression Modelling Strategies book.

Default function behaviour assumes input is a list/data.frame with required vectors as elements.

Usage

calibration(x, ...)

## Default S3 method:
calibration(x, ppid = "prediction", ooid = "reference", gid = "label", ...)

## S3 method for class 'performance'
calibration(x, ...)

## S3 method for class 'dCVnet'
calibration(x, ...)

Arguments

x	an object containing predicted probabilities and observed outcomes, from which calibration can be extracted.
...	arguments to pass on
ppid	indicator for predicted probability element in x (column "name" or index)
ooid	indicator for observed outcome element in x (column "name" or index)
gid	indicator for grouping variable in x (column "name" or index). Set to NA to force no grouping.

Details

Calibration for binomial families is not returned by default for performance because of its computational overhead.

Value

calibration intercept and calibration slope

Methods (by class)

• calibration(performance): binomial calibration for performance objects

• calibration(dCVnet): binomial calibration for dCVnet performance

AndrewLawrence/dCVnet documentation built on Sept. 24, 2024, 5:24 a.m.