def avg_correlation(ts_files, thr=None):
import numpy as np
import numexpr as ne
import hcp_corr
# make empty avg corr matrix
get_size = np.load(ts_files[0]).shape[0]
full_shape = (get_size, get_size)
if np.mod((get_size**2-get_size),2)==0.0:
avg_corr = np.zeros((get_size**2-get_size)/2)
else:
print 'size calculation no zero mod'
count = 0
for ts in ts_files:
# load time series
print '...load %s'%ts
rest = np.load(ts)
# calculate correlations matrix
print '...corrcoef'
corr = hcp_corr.corrcoef_upper(rest)
# corr = np.corrcocoef(rest)
del rest
# get upper triangular only
# corr = corr[np.triu_indices_from(corr, k=1)]
# threshold / transform
if thr == None:
# r-to-z transform and add to avg
print '...transform'
avg_corr += ne.evaluate('arctanh(corr)')
else:
# threshold and add to avg
print '...threshold'
thr = np.percentile(corr, 100-thr)
avg_corr[np.where(corr>thr)] += 1
del corr
count += 1
# divide by number of sessions included
print '...divide'
avg_corr /= count
# transform back if necessary
if thr == None:
print '...back transform'
avg_corr = np.nan_to_num(ne.evaluate('tanh(avg_corr)'))
return avg_corr, full_shape
def avg_correlation(ts_files, thr=None):
'''
Calculates average connectivity matrix using hcp_corr package for memory
optimization: https://github.com/NeuroanatomyAndConnectivity/hcp_corr
'''
# make empty avg corr matrix
if type(ts_files[0]) == str:
get_size = np.load(ts_files[0]).shape[0]
elif type(ts_files[0]) == np.ndarray:
get_size = ts_files[0].shape[0]
full_shape = (get_size, get_size)
if np.mod((get_size**2 - get_size), 2) == 0.0:
avg_corr = np.zeros((get_size**2 - get_size) / 2)
else:
print 'size calculation no zero mod'
count = 0
for rest in ts_files:
# load time series
if type(rest) == str:
rest = np.load(rest)
elif type(rest) == np.ndarray:
pass
# calculate correlations matrix
print '...corrcoef'
corr = hcp_corr.corrcoef_upper(rest)
del rest
# threshold / transform
if thr == None:
# r-to-z transform and add to avg
print '...transform'
avg_corr += ne.evaluate('arctanh(corr)')
else:
# threshold and add to avg
print '...threshold'
thr = np.percentile(corr, 100 - thr)
avg_corr[np.where(corr > thr)] += 1
del corr
count += 1
# divide by number of sessions included
print '...divide'
avg_corr /= count
# transform back if necessary
if thr == None:
print '...back transform'
avg_corr = np.nan_to_num(ne.evaluate('tanh(avg_corr)'))
return avg_corr, full_shape
def avg_correlation(ts_files, thr=None):
'''
Calculates average connectivity matrix using hcp_corr package for memory
optimization: https://github.com/NeuroanatomyAndConnectivity/hcp_corr
'''
# make empty avg corr matrix
if type(ts_files[0])==str:
get_size = np.load(ts_files[0]).shape[0]
elif type(ts_files[0])==np.ndarray:
get_size = ts_files[0].shape[0]
full_shape = (get_size, get_size)
if np.mod((get_size**2-get_size),2)==0.0:
avg_corr = np.zeros((get_size**2-get_size)/2)
else:
print 'size calculation no zero mod'
count = 0
for rest in ts_files:
# load time series
if type(rest)==str:
rest = np.load(rest)
elif type(rest)==np.ndarray:
pass
# calculate correlations matrix
print '...corrcoef'
corr = hcp_corr.corrcoef_upper(rest)
del rest
# threshold / transform
if thr == None:
# r-to-z transform and add to avg
print '...transform'
avg_corr += ne.evaluate('arctanh(corr)')
else:
# threshold and add to avg
print '...threshold'
thr = np.percentile(corr, 100-thr)
avg_corr[np.where(corr>thr)] += 1
del corr
count += 1
# divide by number of sessions included
print '...divide'
avg_corr /= count
# transform back if necessary
if thr == None:
print '...back transform'
avg_corr = np.nan_to_num(ne.evaluate('tanh(avg_corr)'))
return avg_corr, full_shape
def avg_correlation(ts_files, thr=None):
'''
Calculates average connectivity matrix using hcp_corr package for memory
optimization: https://github.com/NeuroanatomyAndConnectivity/hcp_corr
'''
# make empty avg corr matrix
img0 = np.load(ts_files[0])
get_size = img0.shape[1]
del img0
full_shape = (get_size, get_size)
if np.mod((get_size**2-get_size), 2) == 0.0:
avg_corr = np.zeros(int((get_size**2-get_size)/2))
else:
print('size calculation no zero mod')
count = 0
for rest in ts_files:
# load time series
rest = np.load(rest).T
# calculate correlations matrix
print('...corrcoef')
corr = hcp_corr.corrcoef_upper(rest)
del rest
# threshold / transform
print('...transform')
avg_corr += ne.evaluate('arctanh(corr)')
del corr
count += 1
# divide by number of sessions included
print('...divide')
avg_corr /= count
# transform back if necessary
print('...back transform')
avg_corr = np.nan_to_num(ne.evaluate('tanh(avg_corr)'))
return avg_corr, full_shape
# list of all subjects as numpy array
subject_list = np.array(args.subject) # e.g. /ptmp/sbayrak/hcp/*
N = len(subject_list)
for i in range(0, N):
subject = subject_list[i]
print "do loop %d/%d, %s" % (i+1, N, subject)
# load time-series matrix of the subject
K = hcp_corr.t_series(subject, hemisphere=args.hem, N_first=args.N_first,
N_cnt=args.N_cnt)
# get upper-triangular of correlation matrix of time-series as 1D array
K = hcp_corr.corrcoef_upper(K)
print "corrcoef data upper triangular shape: ", K.shape
# # get the full corr matrix
# N_orig = hcp_corr.N_original(K)
# K.resize([N_orig, N_orig])
# hcp_corr.upper_to_down(K)
# print "corrcoef data full matrix shape: ", K.shape
# sum over all subjects
if i == 0:
SUM = K
else:
SUM = ne.evaluate('SUM + K')
t1=np.load(t1_file)
full_shape=tuple((t1.shape[0], t1.shape[0]))
if euclid:
print 'euclid'
t1_3_7_diff = sp.spatial.distance.pdist(t1[:,3:8], 'euclidean')
f = h5py.File(euclid_file%('rh', '3_7'), 'w')
f.create_dataset('upper', data=t1_3_7_diff)
f.create_dataset('shape', data=full_shape)
f.close()
del t1_3_7_diff
if corr:
print 'corr'
t1_3_7_corr = hcp_corr.corrcoef_upper(t1[:,3:8])
f = h5py.File(corr_file%('rh', '3_7'), 'w')
f.create_dataset('upper', data=t1_3_7_corr)
f.create_dataset('shape', data=full_shape)
f.close()
if affinity:
print 'chebychev'
t1_3_7 = t1[:,3:8]
coeff, poly = chebapprox(t1_3_7, degree=4)
#coeff = t1.copy()
print 'affinity'
t1_3_7_affine = dist.compute_affinity(coeff, method=affine_method)
t1_3_7_affine = t1_3_7_affine[np.triu_indices_from(t1_3_7_affine, k=1)]
f = h5py.File(affinity_file, 'w')
f.create_dataset('upper', data=t1_3_7_affine)
matrix = masking.apply_mask(rest, mask)
matrix = matrix.T
cnt_zeros = 0
for i in range(0, matrix.shape[0]):
if np.count_nonzero(matrix[i, :]) == 0:
cnt_zeros += 1
return cnt_zeros, matrix
#### Step 1, get all connectivity matrices of given subject #########
corr_All = []
for image_rest in glob.glob(data_dir + '/' + subject_id + '*' +
'/preprocessed/func/' +
'rest_preprocessed2mni_sm.nii.gz' ):
[voxel_zeros, t_series] = mask_check(image_rest, image_mask)
corr_upper = corrcoef_upper(t_series)
N_orig = N_original(corr_upper)
corr_upper.resize([N_orig, N_orig])
corr = upper_to_down(corr_upper)
print image_rest, corr.shape
corr_All.append(corr)
corr_All = np.array(corr_All)
corr_All = corr_All.T
print 'input data size...', corr_All.shape
##### Step 2, get concordance value per voxel ########################
W_voxels = []
p_voxels = []
Fdist_voxels = []
