#Get similarity matrix all_cc, which is a list of list, where all_cc[i][j] is the correlation coefficient between data i and j. It is an upper trangular matrix.
...
#to use hierarchy module, condensed distance matrix should be available.

nmodel=10000
q=lambda i,j,n: i*n+j-i*(i+1)/2-i-1
condensed_cc=[0]*int(nmodel*(nmodel-1)/2)
for i in range(nmodel-1):
    for j in range(i+1,nmodel):
        index=int(q(i,j,nmodel)):
        condensed_cc[index]=all_cc[i][j]

numpy.save("cc10000",condensed_cc)

condensed_cc=numpy.load("condensed_cc")
#get condensed distance matrix using formula distance=1-c.c.
condensed_dist=1-condensed_cc
#cluster
z=linkage(condensed_dist, 'average', 'correlation')
nc=500 #number of groups
f=fcluster(z,nc,'maxclust')