condvis (version 0.2-2)

ceplot: Interactive conditional expectation plot

Description

Creates an interactive conditional expectation plot, which consists of two main parts. One part is a single plot depicting a section through a fitted model surface, or conditional expectation. The other part shows small data summaries which give the current condition, which can be altered by clicking with the mouse.

Usage

ceplot(data, model, response = NULL, S = NULL, C = NULL, sigma = NULL, 
    distance = "euclidean", type = "default", cex.axis = NULL, cex.lab = NULL, 
    tck = NULL, view3d = FALSE, Corder = "default", selectortype = "minimal", 
    conf = FALSE, select.colour = "blue", select.cex = 1)

Arguments

data

a dataframe containing the training data for the model(s) in model

model

either a single model object, or a list containing model objects.

response

integer index of the response variable in data (character variable name is also acceptable)

S

integer index of the 'free' variable(s) to be used in the expectation plot (character variable name(s) also acceptable). Defaults to the first variable in data that is not the response.

C

a list, each element of which is either an integer index of data or a vector of integer indices of length 2 (character variable names are also acceptable). Defaults to the list produced by

sigma

tuning parameter for colouring data by proximity to current conditioning, passed to visualweight.

distance

distance metric to be used in visualweight, choice of: "euclidean" (default) or "maxnorm"

type

type of interactive plot. "default" plots everything on one device. "separate" plots the condition selector plots on a separate device. "shiny" creates and opens a Shiny web application.

cex.axis

scaling parameter for axis text. Default is 0.7.

cex.lab

scaling parameter for axis labels. Default is 0.8.

tck

parameter for axis ticks. Default is -0.05.

view3d

logical. If TRUE, displays the model section as a prespective plot if possible.

Corder

method of arranging variables in C if C is not supplied. See arrangeC

selectortype

only used if type = "separate". "minimal" is the default behaviour, simply calling plotxc on the variables given by C. "pcp" gives a parallel coo

conf

if TRUE, approximate 95% confidence intervals are plotted where possible

select.colour

colour for highlighting current section in condition selector plots

select.cex

scaling used for plot symbols in condition selector plots

Value

• Returns no value, but produces an interactive plot as a side-effect. The current Rsession is busy until the interactive session is ended by the user, or the graphics device is closed. If using the Shiny option, "Esc" can be pressed to free up the Rsession.

See Also

getGraphicsEvent

Examples

## Example 1: Multivariate regression, xs one continuous predictor

mtcars$cyl <- as.factor(mtcars$cyl)

library(mgcv)
model1 <- list(
    quadratic = lm(mpg ~ cyl + hp + wt + I(wt^2), data = mtcars),
    additive = mgcv::gam(mpg ~ cyl + hp + s(wt), data = mtcars))

conditionvars1 <- list(c("cyl", "hp"))

ceplot(data = mtcars, model = model1, response = "mpg", S = "wt", 
    C = conditionvars1, sigma = 0.3, conf = T)

## Example 2: Binary classification, xs one categorical predictor

mtcars$cyl <- as.factor(mtcars$cyl)
mtcars$am <- as.factor(mtcars$am)

library(e1071)
model2 <- list(
    svm = svm(am ~ mpg + wt + cyl, data = mtcars, family = "binomial"),
    glm = glm(am ~ mpg + wt + cyl, data = mtcars, family = "binomial"))

ceplot(data = mtcars, model = model2, S = "wt", sigma = 1, type = "shiny")

## Example 3: Multivariate regression, xs both continuous

mtcars$cyl <- as.factor(mtcars$cyl)
mtcars$gear <- as.factor(mtcars$gear)

library(e1071)
model3 <- list(svm(mpg ~ wt + qsec + cyl + hp + gear, 
    data = mtcars, family = "binomial"))

conditionvars3 <- list(c("cyl","gear"), "hp")

ceplot(data = mtcars, model = model3, S = c("wt", "qsec"), 
    sigma = 1, C = conditionvars3)

ceplot(data = mtcars, model = model3, S = c("wt", "qsec"), 
    sigma = 1, type = "separate", view3d = T)

## Example 4: Multi-class classification, xs both categorical

mtcars$cyl <- as.factor(mtcars$cyl)
mtcars$vs <- as.factor(mtcars$vs)
mtcars$am <- as.factor(mtcars$am)
mtcars$gear <- as.factor(mtcars$gear)
mtcars$carb <- as.factor(mtcars$carb)

library(e1071)
model4 <- list(svm(carb ~ ., data = mtcars, family = "binomial"))

ceplot(data = mtcars, model = model4, S = c("cyl", "gear"), sigma = 3)

## Example 5: Multi-class classification, xs both continuous

data(wine)
wine$Class <- as.factor(wine$Class)
library(e1071)

model5 <- list(svm(Class ~ ., data = wine))

ceplot(data = wine, model = model5, S = c("Hue", "Flavanoids"), sigma = 3)

ceplot(data = wine, model = model5, S = c("Hue", "Flavanoids"), sigma = 3, 
    type = "separate")

ceplot(data = wine, model = model5, S = c("Hue", "Flavanoids"), sigma = 3, 
    type = "separate", selectortype = "pcp")

## Example 6: Multi-class classification, xs with one categorical predictor,
##            and one continuous predictor.

mtcars$cyl <- as.factor(mtcars$cyl)
mtcars$carb <- as.factor(mtcars$carb)

library(e1071)
model6 <- list(svm(cyl ~ carb + wt + hp, data = mtcars, family = "binomial"))

ceplot(data = mtcars, model = model6, sigma = 1, S = c("carb", "wt"), C = "hp")