varImp2: Calculate variable importance

In correapvf/caretSDM: Species Distribution Models using caret

Calculate variable importance

Description

A generic method for calculating variable importance for objects produced by train.

Usage

varImp2(model, ...)

## S3 method for class 'train'
varImp2(model, nperm = 1, errorFunction = ci_95, ...)

## S3 method for class 'varImp2'
plot(x, scale = FALSE, plot_errorbar = TRUE, ...)

## S3 method for class 'varImp2'
summary(object, scale = FALSE, ...)

Arguments

model	A model returned by train.
...	ignored
nperm	Number of permutations for each variable.
errorFunction	A function used to calculate errors. Function must accept na.rm. Only used if nperm > 1.
x, object	An object from varImp2.
scale	logical. Return scaled values from 0 to 100?
plot_errorbar	logical. Should plot error bars? Error bars are only plotted if scaled = FALSE.

Details

The process of calculating variable importance is different from caret::varImp(model, useModel = TRUE). Here we use the same process as described in biomod2::variables_importance, but the function will directly accept a caret model. If available, multiple cores are used to compute correlations.

Value

An S3 object of class 'varImp2', including:

• importance - A data.table with variables importance, importance from 0 to 100, and errors across permutations.

• resamples - A data.table with correlations for each permutation.

Examples

## Not run: 
v.obj <- varImp2(model)
summary(v.obj)
summary(v.obj, scale = FALSE)
plot(v.obj)

# for multiple models
v.obj <- varImp2(list(model1, model2, model3), nperm = 25)
plot(v.obj, scale = FALSE)

## End(Not run)

correapvf/caretSDM documentation built on June 2, 2022, 8:29 a.m.