psre (version 0.1.2)

glmImp: Importace Measure for Generalized Linear Models

Description

Calculates importance along the lines of Greenwell et al (2018) using partial dependence plots.

Usage

glmImp(obj, varname, data, level = 0.95, ci_method = c("perc", "norm"), ...)

Value

A data frame of importance measures with optimal bootstrapped confidence intervals.

Arguments

obj

Model object, must be able to use predict(obj, type="terms").

varname

Character string giving the name of the variable whose importance will be calculated.

data

A data frame used to estiamte the model.

level

Cofidence level used for the confidence interval.

ci_method

Character string giving the method for calculating the confidence interval - normal or percentile.

...

Other arguments being passed down to aveEffPlot from the DAMisc package.

References

Greenwell, Brandon M., Bradley C. Boehmke and Andrew J. McCarthy. (2018). “A Simple and Effective Model-Based Variable Importance Measure.” arXiv1805.04755 [stat.ML]

Examples

# \donttest{ 
data(gss)
mod <- glm(childs ~ sei10 + sex + educ + age, 
            data=gss, family=poisson)
g_imp1 <- glmImp(mod, "age", gss)
# }