def test_fast_on_real_data():
from pandas import read_table
from os import path as op
k = 200
subdir = "/home2/data/Projects/CWAS/share/nki/subinfo/40_Set1_N104"
ffile = op.join(subdir, "short_compcor_rois_random_k%04i.txt" % k)
fpaths = read_table(ffile, header=None)
fpath = fpaths.ix[0, 0]
import nibabel as nib
img = nib.load(fpath)
dat = img.get_data()
import numpy as np
from CPAC.cwas.subdist import norm_cols, ncor
norm_dat = norm_cols(dat)
corr_dat = norm_dat.T.dot(norm_dat)
ref = eigenvector_centrality(corr_dat)
comp = fast_eigenvector_centrality(norm_dat)
diff = np.abs(ref - comp).mean() # mean diff
print(diff)
ok_(diff < np.spacing(1e10)) # allow some differences
def test_fast_on_real_data():
from pandas import read_table
from os import path as op
k = 200
subdir = "/home2/data/Projects/CWAS/share/nki/subinfo/40_Set1_N104"
ffile = op.join(subdir, "short_compcor_rois_random_k%04i.txt" % k)
fpaths = read_table(ffile, header=None)
fpath = fpaths.ix[0,0]
import nibabel as nib
img = nib.load(fpath)
dat = img.get_data()
import numpy as np
from CPAC.cwas.subdist import norm_cols, ncor
norm_dat = norm_cols(dat)
corr_dat = norm_dat.T.dot(norm_dat)
ref = eigenvector_centrality(corr_dat)
comp = fast_eigenvector_centrality(norm_dat)
diff = np.abs(ref-comp).mean() # mean diff
print(diff)
ok_(diff < np.spacing(1e10)) # allow some differences
def test_fast_eigenvector_centrality(ntpts=100, nvoxs=1000):
print "testing fast_eigenvector_centrality"
# Simulate Data
import numpy as np
from CPAC.cwas.subdist import norm_cols
# Normalize Random Time-Series Data
m = np.random.random((ntpts,nvoxs))
m = norm_cols(m)
# Correlation Data with Range 0-1
mm = m.T.dot(m) # note that need to generate connectivity matrix here
# Execute
#from CPAC.network_centrality.core import fast_eigenvector_centrality,slow_eigenvector_centrality
ref = eigenvector_centrality(mm, verbose=False) # we need to transform mm to be a distance
comp = fast_eigenvector_centrality(m, verbose=False)
diff = np.abs(ref-comp).mean() # mean diff
print(diff)
ok_(diff < np.spacing(1e2)) # allow minimal difference
def test_fast_eigenvector_centrality(ntpts=100, nvoxs=1000):
print "testing fast_eigenvector_centrality"
# Simulate Data
import numpy as np
from CPAC.cwas.subdist import norm_cols
# Normalize Random Time-Series Data
m = np.random.random((ntpts, nvoxs))
m = norm_cols(m)
# Correlation Data with Range 0-1
mm = m.T.dot(m) # note that need to generate connectivity matrix here
# Execute
#from CPAC.network_centrality.core import fast_eigenvector_centrality,slow_eigenvector_centrality
ref = eigenvector_centrality(
mm, verbose=False) # we need to transform mm to be a distance
comp = fast_eigenvector_centrality(m, verbose=False)
diff = np.abs(ref - comp).mean() # mean diff
print(diff)
ok_(diff < np.spacing(1e2)) # allow minimal difference
def get_centrality_by_thresh(timeseries,
template,
method_option,
weight_options,
threshold,
r_value,
memory_allocated):
"""
Method to calculate degree and eigen vector centrality.
This method takes into consideration the amount of memory
allocated by the user to calculate degree centrality.
Parameters
----------
timeseries_data : numpy array
timeseries of the input subject
template : numpy array
Mask/ROI template for timeseries of subject
method_option : integer
0 - degree centrality calculation, 1 - eigenvector centrality calculation, 2 - lFCD calculation
weight_options : string (list of boolean)
list of two booleans for binarize and weighted options respectively
threshold : float
p-value threshold for the correlation values (ignored if the r_value option is specified)
r_value : float
threshold value in terms of the correlation (this will override the threshold option)
memory_allocated : a string
amount of memory allocated to degree centrality
Returns
-------
out_list : string (list of tuples)
list of tuple containing output name to be used to store nifti image
for centrality and centrality matrix
Raises
------
Exception
"""
import numpy as np
import os
from CPAC.network_centrality import calc_blocksize,\
cluster_data,\
convert_pvalue_to_r,\
degree_centrality,\
eigenvector_centrality
from CPAC.cwas.subdist import norm_cols
try:
# Init variables for use
out_list = []
nvoxs = timeseries.shape[0]
ntpts = timeseries.shape[1]
r_matrix = None # init correlation matrix
calc_degree = False # init degree measure flag to false
calc_eigen = False # init eigen measure flag to false
calc_lfcd= False # init lFCD measure flag to false
# Select which method we're going to perform
if method_option == 0:
calc_degree = True
elif method_option == 1:
calc_eigen = True
elif method_option == 2:
calc_lfcd = True
# Set weighting parameters
out_binarize = weight_options[0]
out_weighted = weight_options[1]
# Calculate the block size (i.e., number of voxels) to compute part of the
# connectivity matrix at once.
if calc_eigen:
# We still use a block size to calculate the whole correlation matrix
# because of issues in numpy that lead to extra memory usage when
# computing the dot product.
# See https://cmi.hackpad.com/Numpy-Memory-Issues-BlV9Pg5nRDM.
block_size = calc_blocksize(timeseries, memory_allocated, include_full_matrix=True)
else:
block_size = calc_blocksize(timeseries, memory_allocated)
if r_value == None:
print "Calculating threshold"
r_value = convert_pvalue_to_r(ntpts, threshold)
print "...%s -> %s" % (threshold, r_value)
print "Setup Intermediates/Outputs"
# Degree matrix init
if calc_degree:
print "...degree"
if out_binarize:
degree_binarize = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('degree_centrality_binarize', degree_binarize))
if out_weighted:
degree_weighted = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('degree_centrality_weighted', degree_weighted))
# Eigen matrix init
if calc_eigen:
print "...eigen"
r_matrix = np.zeros((nvoxs, nvoxs), dtype=timeseries.dtype)
if out_binarize:
eigen_binarize = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('eigenvector_centrality_binarize', eigen_binarize))
if out_weighted:
eigen_weighted = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('eigenvector_centrality_weighted', eigen_weighted))
# lFCD matrix init
if calc_lfcd:
print "...degree"
if out_binarize:
lfcd_binarize = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('lFCD_binarize', lfcd_binarize))
if out_weighted:
lfcd_weighted = np.zeros(nvoxs, dtype=timeseries.dtype)
out_list.append(('lFCD_weighted', lfcd_weighted))
# Normalize the timeseries columns for simple correlation calc via dot product later..
print "Normalize TimeSeries"
timeseries = norm_cols(timeseries.T)
# Init blocking indices for correlation matrix calculation
print "Computing centrality across %i voxels" % nvoxs
i = block_size
j = 0
# Calculate correlation matrix in blocks while loop
while i <= nvoxs:
print "running block ->", i, j
try:
print "...correlating"
corr_matrix = np.dot(timeseries[:,j:i].T, timeseries)
except:
raise Exception("Error in calcuating block wise correlation for the block %i,%i"%(j,i))
if calc_eigen:
print "...storing correlation matrix"
r_matrix[j:i] = corr_matrix
if calc_degree:
if out_binarize:
print "...calculating binarize degree"
degree_centrality(corr_matrix, r_value, method="binarize", out=degree_binarize[j:i])
if out_weighted:
print "...calculating weighted degree"
degree_centrality(corr_matrix, r_value, method="weighted", out=degree_weighted[j:i])
if calc_lfcd:
xyz_a = np.argwhere(template)
krange = corr_matrix.shape[0]
print "...iterating through seeds in block - lfcd"
for k in range (0,krange):
corr_seed = corr_matrix[k,:]
labels = cluster_data(corr_seed,r_value,xyz_a)
seed_label = labels[j+k]
if out_binarize:
if seed_label > 0:
lfcd = np.sum(labels==seed_label)
else:
lfcd = 1
lfcd_binarize[j+k] = lfcd
if out_weighted:
if seed_label > 0:
lfcd = np.sum(corr_seed*(labels==seed_label))
else:
lfcd = 1
lfcd_weighted[j+k] = lfcd
print "...removing temporary correlation matrix"
del corr_matrix
j = i
if i == nvoxs:
break
elif (i+block_size) > nvoxs:
i = nvoxs
else:
i += block_size
# In case there are any zeros in lfcd matrix, set them to 1
if calc_lfcd:
if out_binarize:
lfcd_binarize[np.argwhere(lfcd_binarize == 0)] = 1
if out_weighted:
lfcd_weighted[np.argwhere(lfcd_weighted == 0)] = 1
# Perform eigenvector measures if necessary
try:
if calc_eigen:
if out_binarize:
print "...calculating binarize eigenvector"
eigen_binarize[:] = eigenvector_centrality(r_matrix, r_value, method="binarize").squeeze()
if out_weighted:
print "...calculating weighted eigenvector"
eigen_weighted[:] = eigenvector_centrality(r_matrix, r_value, method="weighted").squeeze()
except Exception:
print "Error in calcuating eigen vector centrality"
raise
if calc_degree:
print "...removing effect of auto-correlation on degree"
degree_binarize[degree_binarize!=0] = degree_binarize[degree_binarize!=0] - 1
degree_weighted[degree_weighted!=0] = degree_weighted[degree_weighted!=0] - 1
return out_list
except Exception:
print "Error in calcuating Centrality"
raise
def get_centrality_by_sparsity(timeseries,
method_option,
weight_options,
threshold,
memory_allocated):
"""
Method to calculate degree and eigen vector centrality
Parameters
----------
timeseries : numpy array
timeseries of the input subject
method_options : string (list of boolean)
list of two booleans for degree and eigen options respectively
weight_options : string (list of boolean)
list of two booleans for binarize and weighted options respectively
threshold : float
sparsity threshold for the correlation values
memory_allocated : a string
amount of memory allocated to degree centrality
Returns
-------
out_list : string (list of tuples)
list of tuple containing output name to be used to store nifti image
for centrality and centrality matrix
Raises
------
Exception
"""
import os
import numpy as np
from CPAC.network_centrality import calc_blocksize,\
convert_sparsity_to_r,\
degree_centrality,\
eigenvector_centrality
from CPAC.cwas.subdist import norm_cols
out_list=[]
try:
# Calculate the block size (i.e., number of voxels) to compute part of the
# connectivity matrix at once.
#
# We still use a block size to calculate the whole correlation matrix
# because of issues in numpy that lead to extra memory usage when
# computing the dot product.
# See https://cmi.hackpad.com/Numpy-Memory-Issues-BlV9Pg5nRDM.
block_size = calc_blocksize(timeseries, memory_allocated, include_full_matrix=True)
nvoxs = timeseries.shape[0]
ntpts = timeseries.shape[1]
calc_degree = False # init degree measure flag to false
calc_eigen = False # init eigen measure flag to false
calc_lfcd= False # init lFCD measure flag to false
# Select which method we're going to perform
if method_option == 0:
calc_degree = True
elif method_option == 1:
calc_eigen = True
elif method_option == 2:
calc_lfcd = True
# Set weighting parameters
out_binarize = weight_options[0]
out_weighted = weight_options[1]
corr_matrix = np.zeros((nvoxs, nvoxs), dtype = timeseries.dtype)
print "Normalize TimeSeries"
timeseries = norm_cols(timeseries.T)
print "Computing centrality across %i voxels" % nvoxs
j = 0
i = block_size
while i <= timeseries.shape[1]:
print "running block ->", i,j
print "...correlating"
np.dot(timeseries[:,j:i].T, timeseries, out=corr_matrix[j:i])
j = i
if i == nvoxs:
break
elif (i+block_size) > nvoxs:
i = nvoxs
else:
i += block_size
print "Calculating threshold"
r_value = convert_sparsity_to_r(corr_matrix, threshold, full_matrix = True)
print "r_value ->", r_value
if calc_degree:
if out_binarize:
print "...calculating binarize degree"
degree_binarize = degree_centrality(corr_matrix, r_value, method="binarize")
out_list.append(('degree_centrality_binarize', degree_binarize))
if out_weighted:
print "...calculating weighted degree"
degree_weighted = degree_centrality(corr_matrix, r_value, method="weighted")
out_list.append(('degree_centrality_weighted', degree_weighted))
if calc_eigen:
if out_binarize:
print "...calculating binarize eigenvector"
eigen_binarize = eigenvector_centrality(corr_matrix, r_value, method="binarize")
out_list.append(('eigenvector_centrality_binarize', eigen_binarize))
if out_weighted:
print "...calculating weighted eigenvector"
eigen_weighted = eigenvector_centrality(corr_matrix, r_value, method="weighted")
out_list.append(('eigenvector_centrality_weighted', eigen_weighted))
except Exception:
print "Error while calculating centrality"
raise
return out_list
