modclust

Model based agglomerative clustering. A hierarchy and BIC (Bayesian Information Criterion) values are calculated for given data vectors.

---

Usage 1

modclust(x, htable, bictable, mflag {, min, max, alpha})

x

data matrix NxM, one data vector with length M per row

htable

hierarchy table (reference used for output)

bictable

BIC table (reference used for output)

mflag

method for distance and BIC calculation

mflag	method
0	Single Linkage
1	Complete Linkage (linaer distances)
2	Complete Linkage (log. distances)

min, max

optional minimum (default=2) and maximum (default=N) number of clusters for BIC calculation; 2 <= min < max <= N

alpha

optional factor for BIC calculation (default=1); 0 <= alpha <= 100

Result 1

On return the hierarchy information is stored in htable (Nx3 matrix) and the BIC values are stored in bictable ((max-min+1)x3 matrix). The return value ibest is the index of the BIC table entry with the highest BIC value (ibtest=imax(bictable[*,2]).

hierarchy table htable: N rows, 3 columns

column 0

index of min. row (from)

column 1

index of min. column (to)

column 2

agglomeration cost (distance)

BIC table bictable: max-min+1 rows, 3 columns

column 0

number of clusters

column 1

log. likelihood

column 2

BIC

---

Usage 2

modclust(htable, nclust)

htable

cluster hierarchy table (Nx3 matrix, see Usage 1).

nclust

number of clusters

Result 2

The created partition table p, which is a vector with N elements containing the group indices. The value p[i] (i=0..N) is the index of the cluster containing the data vector i.

---

Usage 3

modclust(ptable, iclust { x})

ptable

partition table (Nx1 matrix, see Usage 2).

nclust

index of cluster to be extracted

x

input data matrix (NxM matrix, see Usage 1)

Result 3

• if x is supplied: data matrix of all data vectors associated with the cluster iclust
• otherwise: index vector containing the indices of the data vectors associated with cluster iclust

---

See also

haclust, em, [../density|density]], svd

<function list>