cv.test {monoClust}

Cross-Validation Test on MonoClust

Description

Perform cross-validation test for different different number of clusters of Monothetic Clustering.

Usage

cv.test(data, fold = 10L, minnodes = 2L, maxnodes = 10L, ncores = 1L, ...)

Arguments

data

Data set to be partitioned.

fold

Number of folds (k). fold = 1 is the special case, when the function performs a Leave-One-Out Cross-Validation (LOOCV).

minnodes

Minimum number of clusters to be checked.

maxnodes

Maximum number of clusters to be checked.

ncores

Number of CPU cores on the current host. When set to NULL, all available cores are used.

...

Other parameters transferred to MonoClust().

Details

The k-fold cross-validation randomly partitions data into k subsets with equal (or close to equal) sizes. k - 1 subsets are used as the training data set to create a tree with a desired number of leaves and the other subset is used as validation data set to evaluate the predictive performance of the trained tree. The process repeats for each subset as the validating set (m = 1, \ldots, k) and the mean squared difference,

MSE_m=\frac{1}{n_m} \sum_{q=1}^Q\sum_{i \in m} d^2_{euc}(y_{iq}, \hat{y}_{(-i)q}),

is calculated, where \hat{y}_{(-i)q} is the cluster mean on the variable q of the cluster created by the training data where the observed value, y_{iq}, of the validation data set will fall into, and d^2_{euc}(y_{iq}, \hat{y}_{(-i)q}) is the squared Euclidean distance (dissimilarity) between two observations at variable $q$. This process is repeated for the $k$ subsets of the data set and the average of these test errors is the cross-validation-based estimate of the mean squared error of predicting a new observation,

CV_K = \overline{MSE} = \frac{1}{M} \sum_{m=1}^M MSE_m.

Value

A MonoClust.cv class containing a data frame of mean sum of square error and its standard deviation.

Note

This function supports parallel processing with foreach::foreach(). It distributes MonoClust calls to processes.

See Also

plot.cv.MonoClust(), MonoClust(), predict.MonoClust()

Examples

library(cluster)
data(ruspini)

# Leave-one-out cross-validation
cv.test(ruspini, fold = 1, minnodes = 2, maxnodes = 4)

# 5-fold cross-validation
cv.test(ruspini, fold = 5, minnodes = 2, maxnodes = 4)