Extract predictions and class probabilities from train objects
These functions can be used for a single
train object or to loop through a number of
train objects to calculate the
training and test data predictions and class probabilities.
## S3 method for class 'list': predict(object, ...)
## S3 method for class 'train': predict(object, newdata = NULL, type = "raw", ...)
extractPrediction(models, testX = NULL, testY = NULL, unkX = NULL, unkOnly = !is.null(unkX) & is.null(testX), verbose = FALSE)
extractProb(models, testX = NULL, testY = NULL, unkX = NULL, unkOnly = !is.null(unkX) & is.null(testX), verbose = FALSE)
predict.train, an object of class
predict.list, a list of objects of class
- an optional set of data to predict on. If
NULL, then the original training data are used
- either "raw" or "prob", for the number/class predictions or class probabilities, respectively. Class probabilities are not available for all classification models
- a list of objects of the class
train. The objects must have been generated with
fitBest = FALSEand
returnData = TRUE.
- an optional set of data to predict
- an optional outcome corresponding to the data given in
- another optional set of data to predict without known outcomes
- a logical to bypass training and test set predictions. This is useful if speed is needed for unknown samples.
- a logical for printing messages
- additional arguments to be passed to other methods
These functions are wrappers for the specific prediction functions in each modeling package. In each case, the optimal tuning values given in the
tuneValue slot of the
finalModel object are used to predict.
To get simple predictions for a new data set, the
predict function can be used. Limits can be imposed on the range of predictions. See
trainControl for more information.
To get predictions for a series of models at once, a list of
train objects can be passes to the
predict function and a list of model predictions will be returned.
The two extraction functions can be used to get the predictions and observed outcomes at once for the training, test and/or unknown samples at once in a single data frame (instead of a list of just the predictions). These objects can then be passes to
predict.train, a vector of predictions if
type = "raw"or a data frame of class probabilities for
type = "probs". In the latter case, there are columns for each class.
predict.list, a list results. Each element is produced by
extractPrediction, a data frame with columns:
obs the observed training and test data pred predicted values model the type of model used to predict object the names of the objects within
modelsis an un-named list, the values of
objectwill be "Object1", "Object2" and so on
dataType "Training", "Test" or "Unknown" depending on what was specified
extractProb, a data frame. There is a column for each class containing the probabilities. The remaining columns are the same as above (although the
predcolumn is the predicted class)
Kuhn (2008), ``Building Predictive Models in R Using the caret'' (
library(mlbench) data(Satellite) numSamples <- dim(Satellite) 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") predict(knnFit) predict(knnFit, newdata = testX) predict(knnFit, type = "prob") bothModels <- list( knn = knnFit, tree = rpartFit) predict(bothModels) predTargets <- extractPrediction( bothModels, testX = testX, testY = testY, unkX = unkX)