def make_data_field():
nsubj = 1
dimx = 60
dimy = 60
pos = 3*np.array([[ 6, 7],
[10, 10],
[15, 10]])
ampli = np.array([5, 7, 6])
sjitter = 6.0
dataset = simul.make_surrogate_array(nbsubj=nsubj, dimx=dimx,
dimy=dimy,
pos=pos, ampli=ampli, width=10.0)
# step 2 : prepare all the information for the parcellation
nbparcel = 10
ref_dim = (dimx,dimy)
xy = np.array(np.where(dataset[0])).T
nvox = np.size(xy,0)
xyz = np.hstack((xy,np.zeros((nvox,1))))
ldata = np.reshape(dataset,(dimx*dimy,1))
anat_coord = xy
mu = 10.
nn = 18
feature = np.hstack((ldata/np.std(ldata),
mu*anat_coord/np.std(anat_coord)))
g = fg.WeightedGraph(nvox)
g.from_3d_grid(xyz.astype(np.int),nn)
g = ff.Field(nvox,g.edges,g.weights,feature)
return g
def test_bsa_methods():
# generate the data
nbsubj=10
dimx=60
dimy=60
pos = 2*np.array([[ 6, 7],
[10, 10],
[15, 10]])
# make a dataset with a nothing feature
null_ampli = np.array([0, 0, 0])
null_dataset = simul.make_surrogate_array(nbsubj=nbsubj,
dimx=dimx,
dimy=dimy,
pos=pos,
ampli=null_ampli,
width=5.0,
seed=1)
null_betas = np.reshape(null_dataset, (nbsubj, dimx, dimy))
# make a dataset with a something feature
pos_ampli = np.array([5, 7, 6])
pos_dataset = simul.make_surrogate_array(nbsubj=nbsubj,
dimx=dimx,
dimy=dimy,
pos=pos,
ampli=pos_ampli,
width=5.0,
seed=2)
pos_betas = np.reshape(pos_dataset, (nbsubj, dimx, dimy))
# set various parameters
theta = float(st.t.isf(0.01, 100))
dmax = 5./1.5
half_subjs = nbsubj/2
thq = 0.9
smin = 5
# tuple of tuples with each tuple being
# (name_of_method, ths_value, data_set, test_function)
algs_tests = (('simple', half_subjs, null_betas, lambda AF, BF: AF == None),
('dev', half_subjs, null_betas, lambda AF, BF: AF == None),
('ipmi', half_subjs, null_betas, lambda AF, BF: AF == None),
('simple', 1, pos_betas, lambda AF, BF: AF.k>1),
('dev', 1, pos_betas, lambda AF, BF: AF.k>1),
('ipmi', 1, pos_betas, lambda AF, BF: AF.k>1),
)
for name, ths, betas, test_func in algs_tests:
# run the algo
AF, BF = make_bsa_2d(betas, theta, dmax, ths, thq, smin, method = name)
yield assert_true, test_func(AF, BF)
def make_dataset(ampli_factor=1.0, nsubj=10):
"""
Generate a standard multi-subject as a set of multi-subject 2D maps
if null, no activation is added
"""
nsubj = 10
dimx = 60
dimy = 60
pos = 2*np.array([[ 6, 7], [10, 10], [15, 10]])
ampli = ampli_factor*np.array([5, 6, 7])
dataset = simul.make_surrogate_array(nbsubj=nsubj, dimx=dimx, dimy=dimy,
pos=pos, ampli=ampli, width=5.0, seed=1)
return dataset
def make_surrogate_data():
"""
"""
dimx = 40
dimy = 40
pos = np.array([[ 2, 10],
[10, 4],
[20, 30],
[30, 20]])
ampli = np.array([5,5,5,5])
data = make_surrogate_array(nbsubj=1, pos=pos, dimx=40, noise_level=0,
dimy=40, ampli=ampli, spatial_jitter=0,
signal_jitter=0).squeeze()
return data
def test_parcel_multi_subj():
"""
"""
# step 1: generate some synthetic data
nsubj = 10
dimx = 60
dimy = 60
pos = 3*np.array([[ 6, 7],
[10, 10],
[15, 10]])
ampli = np.array([5, 7, 6])
sjitter = 6.0
dataset = simul.make_surrogate_array(nbsubj=nsubj, dimx=dimx,
dimy=dimy,
pos=pos, ampli=ampli, width=10.0)
# step 2 : prepare all the information for the parcellation
nbparcel = 10
ref_dim = (dimx,dimy)
xy = np.array(np.where(dataset[0])).T
nvox = np.size(xy,0)
xyz = np.hstack((xy,np.zeros((nvox,1))))
ldata = np.reshape(dataset,(nsubj,dimx*dimy,1))
anat_coord = xy
mask = np.ones((nvox,nsubj)).astype('bool')
Pa = fp.Parcellation(nbparcel,xyz,mask-1)
# step 3 : run the algorithm
Pa = hp.hparcel(Pa, ldata, anat_coord, mu = 10.0)
# step 4: look at the results
Label = np.array([np.reshape(Pa.label[:,s],(dimx,dimy))
for s in range(nsubj)])
control = True
for s in range(nsubj):
control *= (np.unique(Label)==np.arange(nbparcel)).all()
assert(control)
################################################################################
# Main script
################################################################################
# generate the data
nsubj = 10
dimx = 60
dimy = 60
pos = 2*np.array([[ 6, 7],
[10, 10],
[15, 10]])
ampli = np.array([5, 7, 6])
sjitter = 1.0
dataset = simul.make_surrogate_array(nbsubj=nsubj, dimx=dimx, dimy=dimy,
pos=pos, ampli=ampli, width=5.0)
betas = np.reshape(dataset, (nsubj, dimx, dimy))
# set various parameters
theta = float(st.t.isf(0.01, 100))
dmax = 5./1.5
ths = 1#nsubj/2
thq = 0.9
verbose = 1
smin = 5
method = 'simple'#'dev'#'ipmi'#'sbf'
# run the algo
AF, BF = make_bsa_2d(betas, theta, dmax, ths, thq, smin, method, verbose=verbose)
mp.show()
import matplotlib
import matplotlib.pylab as mp
import nipy.neurospin.graph.field as ff
import nipy.neurospin.utils.simul_2d_multisubject_fmri_dataset as simul
################################################################################
# data simulation
dimx=60
dimy=60
pos = 2*np.array([[ 6, 7],
[10, 10],
[15, 10]])
ampli = np.array([3, 4, 4])
nbvox = dimx*dimy
x = simul.make_surrogate_array(nbsubj=1, dimx=dimx, dimy=dimy,
pos=pos, ampli=ampli, width=10.0)
x = np.reshape(x, (dimx, dimy, 1))
beta = np.reshape(x, (nbvox, 1))
xyz = np.array(np.where(x))
nbvox = np.size(xyz,1)
th = 2.36
# compute the field structure and perform the watershed
Fbeta = ff.Field(nbvox)
Fbeta.from_3d_grid(xyz.T.astype('i'), 18)
Fbeta.set_field(beta)
idx, depth, major, label = Fbeta.custom_watershed(0,th)
#compute the region-based signal average
import numpy as np
import pylab as pl
from nipy.neurospin.utils.simul_2d_multisubject_fmri_dataset import \
make_surrogate_array
pos = np.array([[10, 10],
[14, 20],
[23, 18]])
ampli = np.array([4, 5, 2])
# First generate some noiseless data
noiseless_data = make_surrogate_array(nbsubj=1, noise_level=0, spatial_jitter=0,
signal_jitter=0, pos=pos, ampli=ampli)
pl.figure(figsize=(10, 3))
pl.subplot(1, 4, 1)
pl.imshow(noiseless_data[0])
pl.title('Noise-less data')
# Second, generate some group data, with default noise parameters
group_data = make_surrogate_array(nbsubj=3, pos=pos, ampli=ampli)
pl.subplot(1, 4, 2)
pl.imshow(group_data[0])
pl.title('Subject 1')
pl.subplot(1, 4, 3)
pl.title('Subject 2')
pl.imshow(group_data[1])
pl.subplot(1, 4, 4)
pl.title('Subject 3')