Test DD-Classifier
Performs a cross-validation procedure over the given data.
On each step every numchunks observation is removed from the data, the DD-classifier is trained on these data and tested on the removed observations.
ddalpha.getErrorRateCV (data, numchunks = 10, ...)
data |
Matrix containing training sample where each of n rows is one object of the training sample where first d entries are inputs and the last entry is output (class label). |
numchunks |
number of subsets of testing data. Equals to the number of times the classifier is trained. |
... |
additional parameters passed to |
errors |
the part of incorrectly classified data |
time |
the mean training time |
time_sd |
the standard deviation of training time |
ddalpha.train to train the DDα-classifier,
ddalpha.classify for classification using DDα-classifier,
ddalpha.test to test the DD-classifier on particular learning and testing data,
ddalpha.getErrorRatePart to perform a benchmark study of the DD-classifier on particular data.
# Generate a bivariate normal location-shift classification task
# containing 200 training objects and 200 to test with
class1 <- mvrnorm(150, c(0,0),
matrix(c(1,1,1,4), nrow = 2, ncol = 2, byrow = TRUE))
class2 <- mvrnorm(150, c(2,2),
matrix(c(1,1,1,4), nrow = 2, ncol = 2, byrow = TRUE))
propertyVars <- c(1:2)
classVar <- 3
data <- rbind(cbind(class1, rep(1, 150)), cbind(class2, rep(2, 150)))
# Train 1st DDalpha-classifier (default settings)
# and get the classification error rate
stat <- ddalpha.getErrorRateCV(data, numchunks = 5)
cat("1. Classification error rate (defaults): ",
stat$error, ".\n", sep = "")
# Train 2nd DDalpha-classifier (zonoid depth, maximum Mahalanobis
# depth classifier with defaults as outsider treatment)
# and get the classification error rate
stat2 <- ddalpha.getErrorRateCV(data, depth = "zonoid",
outsider.methods = "depth.Mahalanobis")
cat("2. Classification error rate (depth.Mahalanobis): ",
stat2$error, ".\n", sep = "")Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.