decision_boundary: Plot decision boundary for a classifier with 2 predictors

In dajmcdon/ubc-stat406-labs: Tutorials and labs for UBC Stat 406 in the 2020-2021 online year

Description

Plot decision boundary for a classifier with 2 predictors

Usage

decision_boundary(
  fitted_model,
  y,
  x1,
  x2,
  predict_type = NULL,
  predict_fun = NULL,
  resolution = 100,
  showgrid = TRUE,
  ...
)

Arguments

fitted_model	a fitted classifier object, for example the result of glm(...)
y	the original observed classes
x1	original first predictor
x2	original second predictor
predict_type	default is NULL, many predict methods accept a type argument, if known, this can help to produce classifications
predict_fun	if your classifier is different than one of the standard methods (see below) you can pass a custom function which accepts a single argument g (a new data dataframe with columns) and produces a vector of classes
resolution	number of points along the two dimensions to produce classes
showgrid	do we plot a grid of predictions
...	additional arguments passed to plot()

Details

This function should work with only the mandatory arguments if the classifier is produced via glm(), MASS::lda(), MASS::qda() or from either multinom() or nnet() in the {nnet} package. Other classifiers likely require additional instructions. If the predict method produces integer or factor labels as output, it should just "work".

Value

If assigned to an object, returns the matrix of predicted classes

Examples

x1 = runif(100,-1,1)
x2 = runif(100,-1,1)
y = rbinom(100, 1, 1/(1+exp(-2*x1-3*x2)))

logit = glm(y~x1+x2, family="binomial")
disc = MASS::lda(y~x1+x2)

par(mfrow=c(1,2))
decision_boundary(logit, y, x1, x2)
decision_boundary(disc, y, x1, x2)
par(mfrow=c(1,1))

dajmcdon/ubc-stat406-labs documentation built on Aug. 18, 2020, 1:23 p.m.