marginal will be an R package for calculating marginal effects for arbitrary prediction models. As of now, code and research are in early stages. In other words, BUYER BEWARE
Open issues include:
- Finding a valid way to calculate confidence intervals
- Handling categorical variables more elegantly
- Finding the best default behaviors for calculations
- And more!
Some basic usage
devtools::install_github("tommyjones/marginal")
library(marginal)
library(randomForest)
data(mtcars)
fit <- randomForest(mpg ~., data = mtcars)
mfx <- CalcMfx(object = fit, X = mtcars)
plot(mfx)