print "loop done"
# get average
SUM = ne.evaluate('SUM / N')
# Fisher z to r transform on average , now this is back to correlation array
SUM = fisher_z2r(SUM)
# transform correlation array into similarity array
SUM += 1.0
SUM /= 2.0
# get full similartiy matrix of correlations
N = hcp_util.N_original(SUM)
SUM.resize([N,N])
hcp_util.upper_to_down(SUM)
print "SUM.shape", SUM.shape
print "do embed for correlation matrix:", SUM.shape
# Satra's embedding algorithm
embedding, result = embed.compute_diffusion_map(SUM, alpha=0, n_components=20,
diffusion_time=0, skip_checks=True, overwrite=True)
# output prefix
out_prfx=cliarg_out_prfx
# output precision
out_prec="%g"
np.savetxt(out_prfx + "embedding2.csv", embedding, fmt=out_prec, delimiter='\t', newline='\n')
for idx in range(x.shape[0]-1, -1, -1):
back_sum += x[idx]/float(x.sum())
if back_sum >= ten_percent:
thr = bins[idx]
print "top-10percent threshold:", thr
break
# binarize K via thresholding
K[np.where( K >= thr) ] = 1.0
K[np.where( K < thr) ] = 0
elif args.histogram == "node":
# find a threshold value for each row of corr matrix
# convert upper-triangular to full matrix
N_orig = hcp_util.N_original(K)
K.resize([N_orig, N_orig])
hcp_util.upper_to_down(K)
dbins = 0.1
bins = np.arange(-1, 1+dbins, dbins)
for j in range(0, N_orig):
x, bins = np.histogram(K[j,:], bins)
back_sum = 0
for idx in range(x.shape[0]-1, -1, -1):
back_sum += x[idx]/float(x.sum())
if back_sum >= ten_percent:
thr = bins[idx]
#print "top-10percent node threshold:", thr
break
# binarize corr matrix via thresholding
K[j,:][np.where( K[j,:] >= thr) ] = 1.0
K[j,:][np.where( K[j,:] < thr) ] = 0