def testvoronoi(self):
X = nr.randn(10000,2)
A = np.concatenate([np.ones((7000,2)),np.zeros((3000,2))])
X = X+3*A
C = np.array([[0,0],[3,3]])
L = fc.voronoi(X,C)
l = L[:7000].astype('d')
self.assert_(np.mean(l,0)>0.5)
def hparcel(Pa,ldata,anat_coord,nbperm=0,niter=5, mu=10.,dmax = 10., lamb = 100.0, chunksize = 1.e5,verbose=0):
"""
Function that performs the parcellation by optimizing the
sinter-subject similarity while retaining the connectedness
within subject and some consistency across subjects.
Parameters
----------
Pa: a Parcel structure that essentially contains
the grid position information and the individual masks
anat_coord: array of shape(nbvox,3) which defines the position
of the grid points in some space
nbperm=0: the number of times the parcellation and prfx
computation is performed on sign-swaped data
niter=10: number of iterations to obtain the convergence of the method
information in the clustering algorithm
mu=10., float, relative weight of anatomical information
Results
-------
Pa: the resulting parcellation structure appended with the labelling
"""
# a various parameters
nn = 18
nbvox = Pa.nbvox
xyz = Pa.ijk
Sess = Pa.nb_subj
Gs = []
Feature = []
RFeature = []
Ranat = []
# browse the data
for s in range(Sess):
lxyz = xyz[Pa.label[:,s]>-1].astype(np.int)
lnvox = np.sum(Pa.label[:,s]>-1)
lac = anat_coord[Pa.label[:,s]>-1]
g = fg.WeightedGraph(lnvox)
g.from_3d_grid(lxyz,nn)
g.remove_trivial_edges()
beta = np.reshape(ldata[s],(lnvox,ldata[s].shape[1]))
feature = np.hstack((beta,mu*lac/(1.e-15+np.std(anat_coord,0))))
Gs.append(g)
Feature.append(feature)
aux = np.argsort(rand(lnvox))[:np.minimum(chunksize/Sess,lnvox)]
RFeature.append(feature[aux,:])
Ranat.append(lac[aux,:])
RFeature = np.concatenate(RFeature)
Ranat = np.concatenate(Ranat)
# main function
U,proto_anat = optim_hparcel(Ranat, RFeature, Feature, Pa, Gs, anat_coord,
lamb, dmax, niter=niter,
verbose=verbose)
# write the individual labelling
Labels = -1*np.ones((nbvox,Sess)).astype(np.int)
for s in range(Sess):
Labels[Pa.label[:,s]>-1,s] = U[s]
palc = Pa.label.copy()
Pa.set_labels(Labels)
if Pa.isfield('functional'): Pa.remove_feature('functional')
Pa.make_feature(ldata,'functional')
prfx = Pa.PRFX('functional')
# compute the group-level labels
u = fc.voronoi(anat_coord,proto_anat)
Pa.group_labels = u
if nbperm>0:
Pb = fp.Parcellation(Pa.k,xyz,palc)
prfx0 = perm_prfx(Pb,Gs,Feature, ldata,anat_coord,nbperm,niter,dmax, lamb, chunksize)
return Pa,prfx0
else:
return Pa
