Clustering Evaluation Index Based on Leave-one-out One-nearest-neighbor Evaluation

Share:

Description

Computes the leave-one-out one-nearest-neighbor cross-validation of an arbitrary distance matrix.

Usage

1
loo1nn.cv(d, G)

Arguments

d

A dist object.

G

Integer vector with the labels of the true cluster solution. Each element of the vector specifies the cluster 'id' that the element belongs to.

Details

Computes the proportion of succesful clusters that the given distance matrix produces using leave-one-out one-nearest-neighbor cross-validation. Distance ties are solved by majority vote. A tie while voting produces a warning and is solved by selecting a candidate cluster at random.

Value

The computed proportion.

Author(s)

Pablo Montero Manso, José Antonio Vilar.

See Also

cluster.evaluation, loo1nn, knn.cv,

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
 data(synthetic.tseries)
 
 #create the ground thruth cluster 
 G <- rep(1:6, each = 3)
 
 #obtain candidate distance matrix (dist object)
 dACF <- diss(synthetic.tseries, "ACF")
 
 #calculate the cross-validation
 loo1nn.cv(dACF, G)
 

Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.