Question

I have a distance matrix presents the distance matrix for pairwise elements
such as

    A B C D .....
A   n1 n2 n3
B n1    
C n2 n4
D n3 n5 ....... 
E.........

i input the array like for clustering

 arry=  [ 0 n1, n2, n3..
   n1.......
   n2 n4
   n3 n5 ]


Y=sch.linkage(arry,'single')
cutoff=1e-6
T=sch.fcluster(Y, cutoff,'distance')
print T

Z=sch.dendrogram(Y, color_threshold=cutoff)

my fcluster output is like
[ 4 10 12 1 5 13 2 11 1 7 8 3 14 6 10 16 9 15 1 7]
from a previous poster of others
Clustering with scipy – clusters via distance matrix, how to get back the original objects

I understand
the output T[i] only presents the number of element in a cluster ..how I link the original element A, B ,C ,D ,E….. elements with the cluster result and the dendrogram? and lab them properly into my figures.

Answer 1

“I understand the output T[i] only presents the number of element in a cluster…”

T[j] is the “cluster number” of the j-th data point. That is, fcluster provides the assignments of data points to clusters. So, for example, if there are five data points, and fcluster puts the first, second and last in cluster 1 and the others in cluster 2, the return value of fcluster will be array([1, 1, 2, 2, 1]).

Here’s a demo that shows how you can pull that data apart. For convenience, I’ve used fclusterdata instead of the combination of linkage and fcluster. fclusterdata returns the same thing as fcluster.

import numpy as np

def cluster_indices(cluster_assignments):
    n = cluster_assignments.max()
    indices = []
    for cluster_number in range(1, n + 1):
        indices.append(np.where(cluster_assignments == cluster_number)[0])
    return indices

if __name__ == "__main__":
    from scipy.cluster.hierarchy import fclusterdata

    # Make some test data.
    data = np.random.rand(15,2)

    # Compute the clusters.
    cutoff = 1.0
    cluster_assignments = fclusterdata(data, cutoff)

    # Print the indices of the data points in each cluster.
    num_clusters = cluster_assignments.max()
    print "%d clusters" % num_clusters
    indices = cluster_indices(cluster_assignments)
    for k, ind in enumerate(indices):
        print "cluster", k + 1, "is", ind

Typical output:

4 clusters
cluster 1 is [ 0  1  6  8 10 13 14]
cluster 2 is [ 3  4  5  7 11 12]
cluster 3 is [9]
cluster 4 is [2]