In bertcarnell/tornado: Plots for Model Sensitivity and Variable Importance

require(knitr)
require(survival)
require(caret)
require(randomForest)
require(tornado)
knitr::opts_chunk$set(fig.path = "man/figures/")

cat(paste0("|<sub>", 
           paste0(
             c("Actions", "Code Coverage", "Website", "Downloads", "CRAN"), 
             collapse="</sub>|<sub>"), 
           "</sub>|\n"))
cat(paste0("|", paste0(rep(":---:", 5), collapse="|"), "|\n"))
cat(paste0("|", 
           paste0(c("[![R-CMD-CHECK](https://github.com/bertcarnell/tornado/actions/workflows/r-cmd-check.yml/badge.svg)](https://github.com/bertcarnell/tornado/actions/workflows/r-cmd-check.yml)",
                    "[![Coverage status](https://codecov.io/gh/bertcarnell/tornado/branch/master/graph/badge.svg)](https://codecov.io/github/bertcarnell/tornado?branch=master)",
                    "[![](https://img.shields.io/badge/pkgdown-tornado-blue.svg)](https://bertcarnell.github.io/tornado/)",
                    "[![](https://cranlogs.r-pkg.org/badges/tornado)](https://cran.r-project.org/package=tornado)",
                    "[![CRAN status](https://www.r-pkg.org/badges/version/tornado)](https://cran.r-project.org/package=tornado)"),
                  collapse="|"), 
           "|"))

tornado

tornado plots for model sensitivity analysis.

website

Installation

You can also install the development version of tornado from github with:

if (!require(devtools)) install.packages("devtools")
devtools::install_github("bertcarnell/tornado")

Quick Start - Torndao Plots

Linear Models

lm1 <- lm(mpg ~ cyl*wt*hp, data = mtcars)
torn1 <- tornado(lm1, type = "PercentChange", alpha = 0.10)
plot(torn1, xlabel = "MPG", geom_bar_control = list(width = 0.5))

Generalized Linear Models

Plots can also be created for the range of the input data

glm2 <- glm(vs ~ wt + disp + cyl, data = mtcars, family = binomial(link = "logit"))
torn2 <- tornado(glm2, type = "ranges", alpha = 0.10)
plot(torn2, xlabel = "Probability of Straight Engine", geom_bar_control = list(width = 0.5))

Plots can include factors and percentiles of the data as well.

mydata <- mtcars
mydata$am <- factor(mtcars$am)
mydata$vs <- factor(mtcars$vs)
glm2 <- glm(mpg ~ wt + cyl + am + vs, data = mydata, family = gaussian)
torn2 <- tornado(glm2, type = "percentiles", alpha = 0.10)
plot(torn2, xlabel = "MPG", 
     geom_bar_control = list(width = 0.5),
     sensitivity_colors = c("#E41A1C", "#377EB8"),
     geom_point_control = list(fill = "red", col = "red", size = 3))

Censored Data

Accelerated Failure Model

survreg3 <- survival::survreg(survival::Surv(futime, fustat) ~ ecog.ps + rx + age + resid.ds, 
                              survival::ovarian, dist = 'weibull', scale = 1)
torn3 <- tornado(survreg3, modeldata = survival::ovarian, type = "PercentChange", alpha = 0.10)
plot(torn3, xlabel = "Survival Time", geom_bar_control = list(width = 0.5))

Cox Proportional Hazard Model with 2 Standard Deviation bars

coxph4 <- survival::coxph(survival::Surv(stop, event) ~ rx + size + number,
                          survival::bladder)
torn4 <- tornado(coxph4, modeldata = survival::bladder, type = "StdDev",
                 alpha = 2)
plot(torn4, xlabel = "Risk", geom_bar_control = list(width = 0.5))

Machine Learning Models from caret::train

The plot method can also return a ggplot object

mydat <- mtcars
mydat$am <- factor(mydat$am, labels = c("A", "M"))
mydat$vs <- factor(mydat$vs, labels = c("V", "S"))
mydat$cyl <- factor(mydat$cyl)
mydat$gear <- factor(mydat$gear)
mydat$carb <- factor(mydat$carb)

gtest <- caret::train(x = subset(mydat, select = -vs), y = mydat$vs, method = "rf")
torn <- tornado(gtest, type = "percentiles", alpha = 0.10, class_number = 1)
g <- plot(torn, plot = FALSE, xlabel = "Probability of Class V-shaped Engine", geom_bar_control = list(width = 0.5))
g <- g + ggtitle("Classifier caret::train randomforest, 90-110% of the mean in each variable")
plot(g)