Plot Observed versus Predicted Results in Regression and Classification Models

This function takes an object (preferably from the function extractPrediction) and creates a lattice plot. For numeric outcomes, the observed and predicted data are plotted with a 45 degree reference line and a smoothed fit. For factor outcomes, a dotplot plot is produced with the accuracies for the different models.

plotObsVsPred(object, equalRanges = TRUE, ...)

Arguments

object	an object (preferably from the function extractPrediction. There should be columns named obs, pred, model (e.g. "rpart", "nnet" etc.) and dataType (e.g. "Training", "Test" etc)
equalRanges	a logical; should the x- and y-axis ranges be the same?
...	parameters to pass to xyplot or dotplot, such as auto.key

Value

A lattice object. Note that the plot has to be printed to be displayed (especially in a loop).

Details

If the call to extractPrediction included test data, these data are shown, but if unknowns were also included, they are not plotted

Examples

if (FALSE) {
# regression example
data(BostonHousing)
rpartFit <- train(BostonHousing[1:100, -c(4, 14)],
                  BostonHousing$medv[1:100],
                  "rpart", tuneLength = 9)
plsFit <- train(BostonHousing[1:100, -c(4, 14)],
                BostonHousing$medv[1:100],
                "pls")

predVals <- extractPrediction(list(rpartFit, plsFit),
                              testX = BostonHousing[101:200, -c(4, 14)],
                              testY = BostonHousing$medv[101:200],
                              unkX = BostonHousing[201:300, -c(4, 14)])

plotObsVsPred(predVals)


#classification example
data(Satellite)
numSamples <- dim(Satellite)[1]
set.seed(716)

varIndex <- 1:numSamples

trainSamples <- sample(varIndex, 150)

varIndex <- (1:numSamples)[-trainSamples]
testSamples <- sample(varIndex, 100)

varIndex <- (1:numSamples)[-c(testSamples, trainSamples)]
unkSamples <- sample(varIndex, 50)

trainX <- Satellite[trainSamples, -37]
trainY <- Satellite[trainSamples, 37]

testX <- Satellite[testSamples, -37]
testY <- Satellite[testSamples, 37]

unkX <- Satellite[unkSamples, -37]

knnFit  <- train(trainX, trainY, "knn")
rpartFit <- train(trainX, trainY, "rpart")

predTargets <- extractPrediction(list(knnFit, rpartFit),
                                 testX = testX,
                                 testY = testY,
                                 unkX = unkX)

plotObsVsPred(predTargets)
}