bootcluster (version 0.4.2)

k.select_ref: Estimate number of clusters

Description

Estimate number of clusters by bootstrapping stability

Usage

k.select_ref(df, k_range = 2:7, n_ref = 5, B = 100, B_ref = 50, r = 5)

Value

profile

vector of ( Smin_diff(k) - ( Smin_diff(k+1) + se(Smin_diff(k+1)) ) ) measures for researchers's inspection

k

estimated number of clusters

Arguments

df

data.frame of the input dataset

k_range

integer valued vector of the numbers of clusters k to be tested upon

n_ref

number of reference distribution to be generated

B

number of bootstrap re-samples

B_ref

number of bootstrap resamples for the reference distributions

r

number of runs of k-means

Author

Tianmou Liu

Details

This function uses the out-of-bag scheme to estimate the number of clusters in a dataset. The function calculate the Smin of the dataset and at the same time, generate a reference dataset with the same range as the original dataset in each dimension and calculate the Smin_ref. The differences between Smin and Smin_ref at each k,Smin_diff(k), is taken into consideration as well as the standard deviation of the differences. We choose the k to be the argmax of ( Smin_diff(k) - ( Smin_diff(k+1) + (Smin_diff(k+1)) ) ). If Smin_diff(k) less than 0.1 for all k in k_range, we say k = 1

References

Bootstrapping estimates of stability for clusters, observations and model selection. Han Yu, Brian Chapman, Arianna DiFlorio, Ellen Eischen, David Gotz, Matthews Jacob and Rachael Hageman Blair.

Examples

# \donttest{
set.seed(1)
data(iris)
df <- data.frame(iris[,1:4])
df <- scale(df)
k.select_ref(df, k_range = 2:7, n_ref = 5, B=500, B_ref = 500, r=5)
# }