mplot3_decision: 'mplot3': Decision boundaries

In egenn/rtemis: Machine Learning and Visualization

mplot3: Decision boundaries

Description

Plot classification decision boundaries of rtemis models

Usage

mplot3_decision(
  rtmod,
  data,
  vars = c(1, 2),
  dots.per.axis = 100,
  bg.cex = 0.5,
  bg.alpha = 0.4,
  bg.pch = 15,
  par.reset = TRUE,
  theme = "white",
  col = c("#18A3AC", "#F48024"),
  contour.col = "black",
  contour.lwd = 0.1,
  point.pch = c(3, 4),
  point.alpha = 1
)

Arguments

rtmod	rtemics trained model
data	Matrix / data frame of features; last column is class
vars	Integer vector, length 2: Index of features (columns of x) to use to draw decision boundaries. Default = c(1, 2)
dots.per.axis	Integer: Draw a grid with this many dots on each axis. Default = 100
bg.cex	Float: Point cex for background / decision surface. Default = .5
bg.alpha	Float: Point alpha for background / decision surface. Default = .2
bg.pch	Integer vector: pch for background / decision surface. Default = c(3, 4)
par.reset	Logical: If TRUE, reset par before exiting. Default = TRUE
theme	Character: Theme for mplot3_xy, "light" or "dark". Default = "light'
col	Color vector for classes. Default = ucsfPalette
contour.col	Color for decision boundary. Default = "black"
contour.lwd	Float: Line width for decision boundary. Default = .3
point.pch	Integer: pch for data points. Default = c(3, 4)
point.alpha	Float: Alpha for data points. Default = 1

Details

If data has more than 2 variables, any variable not selected using vars will be fixed to their mean Underlying model (e.g. randomForest, rpart, etc) must support standard R predict format for classification: predict(model, newdata, type = "class")

Value

Predicted labels for background grid (invisibly)

Author(s)

E.D. Gennatas

Examples

## Not run: 
dat <- as.data.frame(mlbench::mlbench.2dnormals(200))
mod.cart <- s_CART(dat)
mod.rf <- s_RF(dat)
mplot3_decision(mod.cart, dat)
mplot3_decision(mod.rf, dat)

## End(Not run)

egenn/rtemis documentation built on Nov. 22, 2024, 4:12 a.m.