# load required packages
library(midr)
library(ISLR2)
library(ggplot2)
library(patchwork)
theme_set(theme_midr("y"))
# train a predictive MID model
mid <- interpret(
cbind(bikers, casual, registered) ~ (
mnth + hr + factor(workingday) + weathersit + temp + hum + windspeed
)^2 # model formula
, link = "log"
, data = Bikeshare # training data
, lambda = .05 # smoothing parameter
, subset = casual * registered > 0
)
#> 'model' not passed: response variable in 'data' is used
Feature Importance
# compute MID based variable importance
imp <- mid.importance(mid)
# create importance plots
p1 <- ggmid(imp[[1]], type = "box", max.nterms = 10) +
labs(subtitle = "Effect on Total Bike Rentals (log-scale)") +
theme(legend.position = "none")
p2 <- ggmid(imp[[2]], max.nterms = 10, fill = "maroon", width = .6) +
labs(subtitle = "Casual Bikers") +
theme(legend.position = "none")
p3 <- ggmid(imp[[3]], max.nterms = 10, fill = "steelblue", width = .6) +
labs(subtitle = "Registered Bikers") +
theme(legend.position = "none")
p1 / (p2 + p3)
Component Functions
# create component plots
p1 <- ggmid(mid[[2]], "hr", fill = "maroon", width = .8) +
labs(subtitle = "Main Effect of Hour (Casual Bikers)")
p2 <- ggmid(mid[[2]], "factor(workingday)", fill = "maroon", width = .8) +
labs(subtitle = "Main Effect of Workingday (Casual Bikers)")
p3 <- ggmid(mid[[2]], "temp") +
labs(subtitle = "Main Effect of Temperature (Casual Bikers)")
p4 <- ggmid(mid[[2]], "hr:factor(workingday)", main.effects = TRUE) +
labs(subtitle = "Total Effect of Hour:Workingday (Casual Bikers)") +
theme(legend.position = "none")
(p1 + p2) / (p3 + p4)
# create component plots
p1 <- ggmid(mid[[3]], "hr", fill = "steelblue", width = .8) +
labs(subtitle = "Main Effect of Hour (Registered Bikers)")
p2 <- ggmid(mid[[3]], "mnth", fill = "steelblue", width = .8) +
labs(subtitle = "Main Effect of Month (Registered Bikers)")
p3 <- ggmid(mid[[3]], "temp") +
labs(subtitle = "Main Effect of Temperature (Registered Bikers)")
p4 <- ggmid(
mid[[3]],
"hr:factor(workingday)",
main.effects = TRUE
) +
labs(subtitle = "Total Effect of Hour:Workingday (Registered Bikers)") +
theme(legend.position = "none")
(p1 + p2) / (p3 + p4)
Conditional Expectations
# compute ICE plots
set.seed(42)
ice_rows <- sample(nrow(Bikeshare), 200L)
ice <- mid.conditional(
mid[2:3]
, max.nsamples = 200
, variable = "hr"
, data = Bikeshare[ice_rows, ]
)
# create plots
p1 <- ggmid(
ice[[1]]
, theme = "muted"
, var.color = `factor(workingday)`
) +
labs(subtitle = "Casual Bikers") +
ylim(0, 600)
p2 <- ggmid(
ice[[2]]
, theme = "muted"
, var.color = `factor(workingday)`
) +
labs(subtitle = "Registered Bikers") +
ylim(0, 600)
p1 / p2
Prediction Breakdown
# compute MID breakdown for the individual penguins
bd100 <- mid.breakdown(mid[2:3], data = Bikeshare, row = 100)
# create plots
ordered <- mid.terms(bd100[[1]])
p1 <- ggmid(
bd100[[1]]
, fill = "maroon"
, terms = ordered
) +
theme(legend.position = "none") +
labs(subtitle = "Casual Bikers") +
xlim(0, 3)
p2 <- ggmid(
bd100[[2]]
, fill = "steelblue"
, terms = ordered
) +
theme(legend.position = "none") +
scale_y_discrete(labels = NULL) +
labs(subtitle = "Registered Bikers") +
xlim(3, 6)
p1 + p2
