def test_gls_recursive(self):
cmd = ["pyhrf_gls", "-r", self.tmp_dir]
output = check_output(cmd)
pyhrf.verbose(1, "output:")
pyhrf.verbose(1, output)
expected_ouput = """%s:
%s/subject1:
%s/subject1/fmri:
paradigm.csv
%s/subject1/fmri/analysis:
analysis_result_1.nii
analysis_result_2.csv
analysis_summary.txt
%s/subject1/fmri/run1:
bold_scan_[1...3].nii
%s/subject1/fmri/run2:
bold_scan_[1...3].nii
%s/subject1/t1mri:
anatomy.{hdr,img}
""" % (
(self.tmp_dir,) * 7
)
if output != expected_ouput:
raise Exception(
"Output of command %s is not as expected.\n"
"Output is:\n%sExcepted:\n%s" % (" ".join(cmd), output, expected_ouput)
)
def __init__(self, sampled_variables, nb_its_max, obs_pace=1, burnin=.3,
sample_hist_pace=-1, obs_hist_pace=-1,):
self.variables = {}
self.sampled_variables = sampled_variables
for v in sampled_variables:
self.set_variable(v.name, v)
def get_fraction_or_nb(nb, tot):
if nb>0. and nb<1.:
return int(round(tot * nb))
else:
return nb
self.nb_its_max = nb_its_max
self.burnin = get_fraction_or_nb(burnin, nb_its_max)
self.smpl_hist_pace = get_fraction_or_nb(sample_hist_pace, nb_its_max)
self.obs_hist_pace = get_fraction_or_nb(obs_hist_pace, nb_its_max)
self.tracked_quantities = {}
pyhrf.verbose(1, 'GibbsSampler init. Burnin: %d, nb_its_max: %d, '\
'smpl_hist_pace: %d, obs_hist_pace: %d,'\
%(self.burnin, self.nb_its_max, self.smpl_hist_pace,
self.obs_hist_pace))
def computeComponentsApost(self, variables, j, XhtQXh):
sIMixtP = variables[self.samplerEngine.I_MIXT_PARAM]
var = sIMixtP.getCurrentVars()
mean = sIMixtP.getCurrentMeans()
rb = variables[self.samplerEngine.I_NOISE_VAR].currentValue
nrls = self.currentValue
#for j in xrange(self.nbConditions):
gTQgjrb = XhtQXh[:,j]/rb # de taille nbVox
ej = self.varYtilde + repmat(nrls[j,:], self.ny, 1) * self.varXh[:,:,j].swapaxes(0,1)
numpy.divide(diag(dot(self.varXhtQ[:,j,:],ej)), rb, self.varXjhtQjeji)
for c in xrange(self.nbClasses): #ici classe: 0 (inactif) ou 1 (actif)
self.varClassApost[c,j,:] = 1./(1./var[c,j] + gTQgjrb)
numpy.sqrt(self.varClassApost[c,j,:], self.sigClassApost[c,j,:])
if c > 0: # assume 0 stands for inactivating class
numpy.multiply(self.varClassApost[c,j,:],
add(mean[c,j]/var[c,j], self.varXjhtQjeji),
self.meanClassApost[c,j,:])
else:
multiply(self.varClassApost[c,j,:], self.varXjhtQjeji,
self.meanClassApost[c,j,:])
pyhrf.verbose(5, 'meanClassApost %d cond %d :'%(c,j))
pyhrf.verbose.printNdarray(5, self.meanClassApost[c,j,:])
def packSamplerInput(self, roiData):
try:
shrf = self.sampler.getVariable('hrf')
except KeyError:
shrf = self.sampler.getVariable('brf')
hrfDuration = shrf.duration
zc = shrf.zc
simu = None
if simu != None and shrf.sampleFlag==0:
hrfDuration = (len(simu.hrf.get_hrf(0,0))-1)*simu.hrf.dt
pyhrf.verbose(6,'Found simulation data and hrf is '\
'not sampled, setting hrfDuration to:' \
+str(hrfDuration))
pyhrf.verbose(2,'building BOLDSamplerInput ...')
if simu == None or shrf.sampleFlag:
dt = self.dt if (self.dt!=None and self.dt!=0.) else -self.dtMin
elif simu != None and shrf.sampleFlag == 0:
dt = simu.hrf.dt
samplerInput = self.sampler.inputClass(roiData, dt=dt,
typeLFD=self.driftLfdType,
paramLFD=self.driftLfdParam,
hrfZc=zc,
hrfDuration=hrfDuration)
return samplerInput
def _compute_graph(self):
if self.data_type != 'volume':
raise Exception('Can only compute graph for volume data')
pyhrf.verbose(6, 'FmriData._compute_graph() ...')
to_discard = [self.backgroundLabel]
self._graph = parcels_to_graphs(self.roiMask, kerMask3D_6n,
toDiscard=to_discard)
def checkAndSetInitValue(self, variables):
smplVarDrift = variables[self.samplerEngine.I_ETA]
smplVarDrift.checkAndSetInitValue(variables)
varDrift = smplVarDrift.currentValue
if self.useTrueValue :
if self.trueValue is not None:
self.currentValue = self.trueValue
else:
raise Exception('Needed a true value for drift init but '\
'None defined')
if 0 and self.currentValue is None :
#if not self.sampleFlag and self.dataInput.simulData is None :
#self.currentValue = self.dataInput.simulData.drift.lfd
#pyhrf.verbose(6,'drift dimensions :' \
#+str(self.currentValue.shape))
#pyhrf.verbose(6,'self.dimDrift :' \
#+str(self.dimDrift))
#assert self.dimDrift == self.currentValue.shape[0]
#else:
self.currentValue = np.sqrt(varDrift) * \
np.random.randn(self.dimDrift, self.nbVox)
if self.currentValue is None:
pyhrf.verbose(1,"Initialisation of Drift from the data")
ptp = numpy.dot(self.P.transpose(),self.P)
invptp = numpy.linalg.inv(ptp)
invptppt = numpy.dot(invptp, self.P.transpose())
self.currentValue = numpy.dot(invptppt,self.dataInput.varMBY)
self.updateNorm()
self.matPl = dot(self.P, self.currentValue)
self.ones_Q_J = np.ones((self.dimDrift, self.nbVox))
self.ones_Q = np.ones((self.dimDrift))
def remote_copy(files, host, user, path, transfer_tool='ssh'):
if transfer_tool == 'paramiko':
import paramiko
pyhrf.verbose(1, 'Copying files to remote destination %s@%s:%s ...' \
%(host,user,path))
ssh = paramiko.SSHClient()
known_hosts_file = os.path.join("~", ".ssh", "known_hosts")
ssh.load_host_keys(os.path.expanduser(known_hosts_file))
ssh.connect(host, username=user)
sftp = ssh.open_sftp()
for f in files:
remotepath = op.join(path,op.basename(f))
pyhrf.verbose(2, f + ' -> ' + remotepath + ' ...')
flocal = open(f)
remote_file = sftp.file(remotepath, "wb")
remote_file.set_pipelined(True)
remote_file.write(flocal.read())
flocal.close()
remote_file.close()
sftp.close()
ssh.close()
else:
sfiles = string.join(['"%s"'%f for f in files], ' ')
scp_cmd = 'scp -C %s "%s@%s:%s"' %(sfiles, user, host, path)
pyhrf.verbose(1, 'Data files transfer with scp ...')
pyhrf.verbose(2, scp_cmd)
if os.system(scp_cmd) != 0:
raise Exception('Error while scp ...')
pyhrf.verbose(1, 'Copy done!')
return [op.join(path,op.basename(f)) for f in files]
def sampleNextInternal(self, variables):
#TODO : comment
sIMixtP = variables[self.samplerEngine.I_MIXT_PARAM_NRLS_BAR]
varCI = sIMixtP.currentValue[sIMixtP.I_VAR_CI]
varCA = sIMixtP.currentValue[sIMixtP.I_VAR_CA]
meanCA = sIMixtP.currentValue[sIMixtP.I_MEAN_CA]
self.labelsSamples = rand(self.nbConditions, self.nbVox)
self.nrlsSamples = randn(self.nbConditions, self.nbVox)
if self.imm:
#self.sampleNrlsParallel(varXh, rb, h, varLambda, varCI,
# varCA, meanCA, gTQg, variables)
raise NotImplementedError("IMM with drift sampling is not available")
else: #smm
self.sampleNrlsSerial(varCI, varCA, meanCA, variables)
#self.computeVarYTildeOpt(varXh)
#matPl = self.samplerEngine.getVariable('drift').matPl
#self.varYbar = self.varYtilde - matPl
if (self.currentValue >= 1000).any() and \
pyhrf.__usemode__ == pyhrf.DEVEL:
pyhrf.verbose(2, "Weird NRL values detected ! %d/%d" \
%((self.currentValue >= 1000).sum(),
self.nbVox*self.nbConditions) )
#pyhrf.verbose.set_verbosity(6)
if pyhrf.verbose.verbosity >= 4:
self.reportDetection()
self.printState(4)
self.iteration += 1 #TODO : factorize !!
def from_py_object(self, label, obj, parent=None):
pyhrf.verbose(6, "UiNode.from_py_object(label=%s,obj=%s) ..." % (label, str(obj)))
if isinstance(obj, Initable):
n = obj.to_ui_node(label, parent)
else:
if UiNode._pyobj_has_leaf_type(obj):
if isinstance(obj, np.ndarray):
dt = str(obj.dtype.name)
sh = str(obj.shape)
n = UiNode(label, attributes={"type": "ndarray", "dtype": dt, "shape": sh})
s = " ".join(str(e) for e in obj.ravel())
n.add_child(UiNode(s))
elif obj is None:
n = UiNode(label, attributes={"type": "None"})
n.add_child(UiNode("None"))
else:
n = UiNode(label, attributes={"type": obj.__class__})
n.add_child(UiNode(str(obj)))
elif isinstance(obj, list):
n = UiNode(label, attributes={"type": "list"})
for i, sub_val in enumerate(obj):
n.add_child(UiNode.from_py_object("item%d" % i, sub_val))
elif isinstance(obj, (dict, OrderedDict)):
t = ["odict", "dict"][obj.__class__ == dict]
n = UiNode(label, attributes={"type": t})
for k, v in obj.iteritems():
n.add_child(UiNode.from_py_object(k, v))
else:
raise Exception(
"In UiNode.from_py_object, unsupported object: " "%s (type: %s)" % (str(obj), str(type(obj)))
)
return n
def analyse_roi(self, atomData):
"""
Launch the JDE Gibbs Sampler on a parcel-specific data set *atomData*
Args:
- atomData (pyhrf.core.FmriData): parcel-specific data
Returns:
JDE sampler object
"""
#print 'atomData:', atomData
if self.copy_sampler:
sampler = copyModule.deepcopy(self.sampler)
else:
sampler = self.sampler
sInput = self.packSamplerInput(atomData)
sampler.linkToData(sInput)
#if self.parameters[self.P_RANDOM_SEED] is not None:
# np.random.seed(self.parameters[self.P_RANDOM_SEED])
# #HACK:
# if len(self.roi_ids) > 0:
# if atomData.get_roi_id() not in self.roi_ids:
# return None
pyhrf.verbose(1, 'Treating region %d' %(atomData.get_roi_id()))
sampler.runSampling()
pyhrf.verbose(1, 'Cleaning memory ...')
sampler.dataInput.cleanMem()
return sampler
def project_fmri(input_mesh, data_file, output_tex_file,
output_kernels_file=None, data_resolution=None,
geod_decay=5., norm_decay=2., kernel_size=7,
tex_bin_threshold=None):
if output_kernels_file is None:
tmp_dir = tempfile.mkdtemp(prefix='pyhrf_surf_proj',
dir=pyhrf.cfg['global']['tmp_path'])
kernels_file = op.join(tmp_dir, add_suffix(op.basename(data_file),
'_kernels'))
tmp_kernels_file = True
else:
kernels_file = output_kernels_file
tmp_kernels_file = False
if data_resolution is not None:
resolution = data_resolution
else:
resolution = read_spatial_resolution(data_file)
pyhrf.verbose(1,'Data resolution: %s' %resolution)
pyhrf.verbose(2,'Projection parameters:')
pyhrf.verbose(2,' - geodesic decay: %f mm' %geod_decay)
pyhrf.verbose(2,' - normal decay: %f mm' %norm_decay)
pyhrf.verbose(2,' - kernel size: %f voxels' %kernel_size)
create_projection_kernels(input_mesh, kernels_file, resolution,
geod_decay, norm_decay, kernel_size)
project_fmri_from_kernels(input_mesh, kernels_file, data_file,
output_tex_file, tex_bin_threshold)
if tmp_kernels_file:
os.remove(kernels_file)
def test_ward_spatial_real_data(self):
from pyhrf.glm import glm_nipy_from_files
#pyhrf.verbose.verbosity = 2
fn = 'subj0_parcellation.nii.gz'
mask_file = pyhrf.get_data_file_name(fn)
bold = 'subj0_bold_session0.nii.gz'
bold_file = pyhrf.get_data_file_name(bold)
paradigm_csv_file = pyhrf.get_data_file_name('paradigm_loc_av.csv')
output_dir = self.tmp_dir
output_file = op.join(output_dir,
'parcellation_output_test_real_data.nii')
tr=2.4
bet = glm_nipy_from_files(bold_file, tr,
paradigm_csv_file, output_dir,
mask_file, session=0,
contrasts=None,
hrf_model='Canonical',
drift_model='Cosine', hfcut=128,
residuals_model='spherical',
fit_method='ols', fir_delays=[0])[0]
pyhrf.verbose(2, 'betas_files: %s' %' '.join(bet))
cmd = 'pyhrf_parcellate_glm -m %s %s -o %s -v %d -n %d '\
'-t ward_spatial ' \
%(mask_file, ' '.join(bet), output_file,
pyhrf.verbose.verbosity, 10)
if os.system(cmd) != 0 :
raise Exception('"' + cmd + '" did not execute correctly')
pyhrf.verbose(1, 'cmd: %s' %cmd)
def checkAndSetInitHabit(self, variables) :
# init habituation speed factors and time-varying NRLs
if self.habits == None : # if no habituation speed specified
if not self.sampleHabitFlag :
if self.dataInput.simulData != None : # Attention: on a un probleme si on fait tourner ce modele sur des donnees simulees par le modele stationnaire. Dans ce cas, il faut forcer ici a passer et prendre des habits nulles
## using simulated Data for HABITUATION sampling
print "load Habituation from simulData", self.dataInput.simulData.habitspeed.data
self.habits = self.dataInput.simulData.habitspeed.data
self.timeNrls = self.dataInput.simulData.nrls.timeNrls
self.Gamma = self.setupGamma()
else : # sinon, on prend que des zeros (modele stationnaire)
self.habits = numpy.zeros((self.nbConditions, self.nbVox), dtype=float)
self.setupTimeNrls()
else:
pyhrf.verbose(2, "Uniform set up of habituation factors")
habitCurrent = numpy.zeros((self.nbConditions, self.nbVox), dtype=float)
for nc in xrange(self.nbConditions):
habitCurrent[nc,self.voxIdx[1][nc]] = numpy.random.rand(self.cardClass[1][nc])
self.habits = habitCurrent
if self.outputRatio :
self.ratio = zeros((self.nbConditions, self.nbVox, 2), dtype = float)
self.ratiocourbe = zeros((self.nbConditions, self.nbVox, 100, 5), dtype = float)
self.compteur = numpy.zeros((self.nbConditions, self.nbVox), dtype=float)
self.setupTimeNrls()
pyhrf.verbose(4, 'habituation initiale')
pyhrf.verbose.printNdarray(5, self.habits)
def create_hrf_from_territories(hrf_territories, primary_hrfs):
pyhrf.verbose(1,'create_hrf_from_territories ...')
pyhrf.verbose(1,' inputs: hrf_territories %s, primary_hrfs (%d,%d)' \
%(str(hrf_territories.shape), len(primary_hrfs),
len(primary_hrfs[0][0])))
assert hrf_territories.ndim == 1
hrfs = np.zeros((hrf_territories.size,primary_hrfs[0][0].size))
territories = np.unique(hrf_territories)
territories.sort()
if territories.min() == 1:
territories = territories - 1
assert territories.min() == 0
assert territories.max() <= len(primary_hrfs)
#print hrfs.shape
#sm = ','.join(['m[%d]'%d for d in range(hrf_territories.ndim)] + [':'])
for territory in territories:
#TODO: test consitency in hrf lengths
m = np.where(hrf_territories==territory)[0]
# print 'm:',m
# print 'hrfs[m,:].shape:', hrfs[m,:].shape
# print 'primary_hrfs[territory][1].shape:', \
# primary_hrfs[territory][1].shape
# print primary_hrfs[territory][1]
#print hrfs[m,:].shape
#exec('hrfs[%s] = primary_hrfs[territory][1]' %sm)
hrfs[m,:] = primary_hrfs[territory][1]
#print hrfs[m,:]
return hrfs.transpose()
def create_asl_from_stim_induced(bold_stim_induced, perf_stim_induced,
ctrl_tag_mat, dsf,
perf_baseline, noise, drift=None, outliers=None):
"""
Downsample stim_induced signal according to downsampling factor 'dsf' and
add noise and drift (nuisance signals) which has to be at downsampled
temporal resolution.
"""
bold = bold_stim_induced[0:-1:dsf,:].copy()
perf = np.dot(ctrl_tag_mat, (perf_stim_induced[0:-1:dsf,:].copy() + \
perf_baseline))
pyhrf.verbose(3, 'create_asl_from_stim_induced ...')
pyhrf.verbose(3, 'bold shape: %s, perf_shape: %s, noise shape: %s, '\
'drift shape: %s' %(str(bold.shape), str(perf.shape),
str(noise.shape), str(drift.shape)))
asl = bold + perf
if drift is not None:
asl += drift
if outliers is not None:
asl += outliers
asl += noise
return asl
def create_polynomial_drift(bold_shape, tr, drift_order, drift_var):
p = buildPolyMat(drift_order, bold_shape[0], tr)
nvox = bold_shape[1]
coeff = drift_var**.5 * randn(p.shape[1],nvox)
drift = np.dot(p, coeff)
pyhrf.verbose(3, 'Drift shape: %s' %str(drift.shape))
return drift
def updateObsersables(self):
GibbsSamplerVariable.updateObsersables(self)
sHrf = self.samplerEngine.getVariable('hrf')
sScale = self.samplerEngine.getVariable('scale')
if sHrf.sampleFlag and np.allclose(sHrf.normalise,0.) and \
not sScale.sampleFlag and self.sampleFlag:
pyhrf.verbose(6, 'Normalizing Posterior mean of Mixture Parameters at each iteration ...')
#print '%%%% scaling NRL PME %%% - hnorm = ', sHrf.norm
# Undo previous calculation:
self.cumul -= self.currentValue
#self.cumul2 -= self.currentValue**2
self.cumul3 -= (self.currentValue - self.mean)**2
# Use scaled quantities instead:
cur_m_CA = self.currentValue[self.I_MEAN_CA]
cur_v_CA = self.currentValue[self.I_VAR_CA]
cur_v_CI = self.currentValue[self.I_VAR_CI]
self.cumul[self.I_MEAN_CA] += cur_m_CA * sHrf.norm
#self.cumul2[self.I_MEAN_CA] += (cur_m_CA * sHrf.norm)**2
self.cumul[self.I_VAR_CA] += cur_v_CA * sHrf.norm**2
#self.cumul2[self.I_VAR_CA] += (cur_v_CA * sHrf.norm**2)**2
self.cumul[self.I_VAR_CI] += cur_v_CI * sHrf.norm**2
#self.cumul2[self.I_VAR_CI] += (cur_v_CI * sHrf.norm**2)**2
self.mean = self.cumul / self.nbItObservables
self.cumul3[self.I_MEAN_CA] += (cur_m_CA * sHrf.norm - self.mean[self.I_MEAN_CA])**2
self.cumul3[self.I_VAR_CA] += (cur_v_CA * sHrf.norm**2 - self.mean[self.I_VAR_CA])**2
self.cumul3[self.I_VAR_CI] += (cur_v_CI * sHrf.norm**2 - self.mean[self.I_VAR_CI])**2
#self.error = self.cumul2 / self.nbItObservables - \
#self.mean**2
self.error = self.cumul3 / self.nbItObservables
def sampleNrlsSerial(self, rb, h, varCI, varCA, meanCA ,
gTg, variables):
pyhrf.verbose(3, 'Sampling Nrls (serial, spatial prior) ...')
sIMixtP = variables[self.samplerEngine.I_MIXT_PARAM]
var = sIMixtP.getCurrentVars()
mean = sIMixtP.getCurrentMeans()
rb = variables[self.samplerEngine.I_NOISE_VAR].currentValue
# Add one dimension to be consistent with habituation model
varXh = variables[self.samplerEngine.I_HRF].varXh
varXht = varXh.transpose()
nrls = self.currentValue
neighbours = self.dataInput.neighboursIndexes
beta = self.samplerEngine.getVariable('beta').currentValue
voxOrder = permutation(self.nbVox)
sampleSmmNrl2(voxOrder.astype(np.int32), rb.astype(np.float64),
neighbours.astype(np.int32), self.varYbar,
self.labels, np.array([varXh], dtype=np.float64),
self.currentValue,
self.nrlsSamples.astype(np.float64),
self.labelsSamples.astype(np.float64),
np.array([varXht], dtype=np.float64),
gTg.astype(np.float64),
beta.astype(np.float64), mean.astype(np.float64),
var.astype(np.float64), self.meanClassApost,
self.varClassApost, self.nbClasses,
self.sampleLabelsFlag+0, self.iteration,
self.nbConditions)
self.countLabels(self.labels, self.voxIdx, self.cardClass)
def __init__(self, label, parent=None, attributes=None):
super(UiNode, self).__init__()
pyhrf.verbose(6, 'Create new UiNode(%s,%s,%s)' \
%(label, str(parent), str(attributes)))
self._label = label
self._children = []
self._parent = parent
if attributes is None:
attributes = {}
# Check input attributes (has to be a dict of strings):
if not isinstance(attributes, dict):
raise Exception('Wrong type "%s" for attributes, has to be a dict'\
%str(type(attributes)))
for k,v in attributes.iteritems():
if not isinstance(k, str):
raise Exception('Wrong type for attribute "%s" '\
'(has to be string)' %str(k))
if k not in ['init_obj', 'type'] and not isinstance(v, str):
# init_obj can be a method/function (see Initable class)
raise Exception('Wrong type for value of attribute "%s" (%s) '\
'(has to be string)' %(str(k),str(v)))
self._attributes = attributes
if parent is not None:
parent.add_child(self)
def linkToData(self, dataInput):
self.dataInput = dataInput
self.nbConditions = self.dataInput.nbConditions
self.nbVox = self.dataInput.nbVoxels
self.ny = self.dataInput.ny
self.nbColX = self.dataInput.nbColX
# Do some allocations :
self.beta = np.zeros(self.nbVox, dtype=float)
self.mXhQXh = np.zeros((self.nbConditions,self.nbConditions),
dtype=float)
if self.dataInput.simulData is not None:
if isinstance(self.dataInput.simulData, dict):
if dataInput.simulData.has_key('v_gnoise'):
self.trueValue = self.dataInput.simulData['v_noise']
pyhrf.verbose(3, 'True noise variance = %1.3f' \
%self.trueValue)
elif isinstance(dataInput.simulData, list):
sd = dataInput.simulData[0]
if isinstance(sd, dict):
self.trueValue = sd['noise'].var(0).astype(np.float64)
else:
self.trueValue = sd.noise.variance
else:
self.trueValue = self.dataInput.simulData.noise.variance
def split_big_parcels(parcel_file, output_file, max_size=400):
print 'split_big_parcels ...'
roiMask, roiHeader = read_volume(parcel_file)
roiIds = np.unique(roiMask)
background = roiIds.min()
labels = roiMask[np.where(roiMask>background)].astype(int)
if (np.bincount(labels) <= max_size).all():
pyhrf.verbose(1, 'no parcel to split')
return
graphs = parcels_to_graphs(roiMask, kerMask3D_6n)
for roiId in roiIds:
if roiId != background:
roi_size = (roiMask==roiId).sum()
if roi_size > max_size:
print 'roi %d, size = %d' %(roiId, roi_size)
nparcels = int(np.ceil(roi_size*1./max_size))
print 'split into %d parcels ...' %(nparcels)
split_parcel(labels, graphs, roiId, nparcels, inplace=True,
verbosity=1)
final_roi_mask = np.zeros_like(roiMask)
final_roi_mask[np.where(roiMask>background)] = labels
#print np.bincount(labels)
assert (np.bincount(labels) <= max_size).all()
write_volume(final_roi_mask, output_file, roiHeader)
def graph_from_mesh(polygonList):
"""Return the list of neighbours indexes for each position, from a list of
polygons. Each polygon is a triplet.
"""
indexSet = set()
for l in polygonList:
indexSet.update(l)
nbPositions = len(indexSet)
pyhrf.verbose(6, 'indexSet:')
pyhrf.verbose.printDict(6, indexSet)
neighbourSets = {}
#pyhrf.verbose(6, 'polygonList:')
#pyhrf.verbose.printDict(6, polygonList)
for triangle in polygonList:
for idx in triangle:
if not neighbourSets.has_key(idx):
neighbourSets[idx] = set()
neighbourSets[idx].update(triangle)
# idx<->set --> idx<->list
neighbourLists = np.empty(nbPositions, dtype=object)
for idx in indexSet:
neighbourSets[idx].remove(idx)
neighbourLists[idx] = np.array(list(neighbourSets[idx]),dtype=int)
return neighbourLists
def computeWithJeffreyPriors(self, j, cardCDj):
if pyhrf.verbose.verbosity >= 3:
print 'cond %d - card CD = %d' %(j,cardCDj)
print 'cond %d - cur mean CD = %f' %(j,self.currentValue[self.I_MEAN_CD,j])
if cardCDj > 0:
print 'cond %d - nrl CD: %f(v%f)[%f,%f]' %(j,self.nrlCD[j].mean(),
self.nrlCD[j].var(),
self.nrlCD[j].min(),
self.nrlCD[j].max())
if cardCDj > 1:
nrlC2Centered = self.nrlCD[j] - self.currentValue[self.I_MEAN_CD,j]
nu2j = dot(nrlC2Centered, nrlC2Centered)
varCDj = 1.0 / random.gamma(0.5 * (cardCDj + 1) - 1, 2. / nu2j)
eta2j = mean(self.nrlCD[j])
meanCDj = random.normal(eta2j, (varCDj / cardCDj)**0.5)
if pyhrf.verbose.verbosity >= 3:
print 'varCD ~ InvGamma(%f, nu2j/2=%f)' %(0.5*(cardCDj+1)-1,
nu2j/2.)
print ' -> mean =', (nu2j/2.)/(0.5*(cardCDj+1)-1)
else :
pyhrf.verbose(1,'Warning : cardCD <= 1!')
varCDj = 1.0 / random.gamma(.5, 2.)
if cardCDj == 0 :
meanCDj = random.normal(5.0, varCDj**0.5)
else:
meanCDj = random.normal(self.nrlCD[j], varCDj**0.5)
if pyhrf.verbose.verbosity >= 3:
print 'Sampled components - cond', j
print 'mean CD =', meanCDj, 'var CD =', varCDj
return meanCDj, varCDj
def linkToData(self, data):
self.M = data.nbConditions
dp = data.paradigm
self.condition_names = dp.stimOnsets.keys()
self.onsets = dp.stimOnsets.values()
self.LengthOnsets = [dp.stimDurations[n] for n in dp.stimOnsets.keys()]
self.nbVoxels = data.bold.shape[1]
if not self.avg_bold:
self.bold = data.bold
else:
self.bold = data.bold.mean(1)[:,np.newaxis]
pyhrf.verbose(2, 'BOLD is averaged -> shape: %s' \
%str(self.bold.shape))
self.sscans = data.sessionsScans
self.nbSessions = data.nbSessions
self.I = self.nbSessions
self.ImagesNb = sum([len(ss) for ss in self.sscans])
self.TR = data.tr
self.Ni = array([len(ss) for ss in self.sscans])
self.OnsetList = [[o[i] for o in self.onsets] \
for i in xrange(self.nbSessions)]
self.Qi = zeros(self.nbSessions, dtype=int) + 2
self.history = defaultdict(init_dict)
def split(self, dump_sub_results=None, make_sub_outputs=None,
output_dir=None, output_file_list=None):
if dump_sub_results is None:
dump_sub_results = (self.result_dump_file is not None)
if make_sub_outputs is None:
make_sub_outputs = self.make_outputs
if output_dir is None:
output_dir = self.output_dir
sub_treatments = [FMRITreatment(d, deepcopy(self.analyser),
make_outputs=make_sub_outputs,
output_dir=output_dir) \
for d in self.analyser.split_data(self.data)]
if output_dir is not None:
pyhrf.verbose(1, 'Dump sub treatments in: %s ...' %output_dir)
cmp_size = lambda t1,t2:cmp(t1.data.get_nb_vox_in_mask(),
t2.data.get_nb_vox_in_mask())
for it, sub_t in enumerate(sorted(sub_treatments, cmp=cmp_size,
reverse=True)):
if dump_sub_results:
sub_t.result_dump_file = op.join(output_dir,
'result_%04d.pck' %it)
fn = op.join(output_dir, 'treatment_%04d.pck' %it)
fout = open(fn, 'w')
cPickle.dump(sub_t, fout)
fout.close()
if output_file_list is not None:
output_file_list.append(fn)
return sub_treatments
def sampleNrlsSerialWithRelVar(self, rb, h, gTg, variables, w, t1, t2):
pyhrf.verbose(3, 'Sampling Nrls (serial, spatial prior) ...')
sIMixtP = variables[self.samplerEngine.I_MIXT_PARAM]
var = sIMixtP.getCurrentVars()
mean = sIMixtP.getCurrentMeans()
rb = variables[self.samplerEngine.I_NOISE_VAR].currentValue
y = self.dataInput.varMBY
matPl = self.samplerEngine.getVariable('drift').matPl
sampleWFlag = variables[self.samplerEngine.I_W].sampleFlag
# Add one dimension to be consistent with habituation model
varXh = variables[self.samplerEngine.I_HRF].varXh
varXht = varXh.transpose()
nrls = self.currentValue
neighbours = self.dataInput.neighboursIndexes
beta = self.samplerEngine.getVariable('beta').currentValue
voxOrder = permutation(self.nbVox)
cardClassCA = np.zeros(self.nbConditions, dtype=int)
for i in range(self.nbConditions):
cardClassCA[i] = self.cardClass[self.L_CA,i]
#sampleSmmNrl2WithRelVar(voxOrder.astype(np.int32), rb.astype(np.float64),
#neighbours.astype(np.int32), self.varYbar,
#self.labels, np.array([varXh], dtype=np.float64),
#self.currentValue,
#self.nrlsSamples.astype(np.float64),
#self.labelsSamples.astype(np.float64),
#np.array([varXht], dtype=np.float64),
#gTg.astype(np.float64),
#beta.astype(np.float64), mean.astype(np.float64),
#var.astype(np.float64), self.meanClassApost,
#self.varClassApost, w.astype(np.int32), t1, t2,
#cardClassCA.astype(np.int32),
#self.nbClasses,
#self.sampleLabelsFlag+0, self.iteration,
#self.nbConditions)
sampleSmmNrl2WithRelVar_NEW(voxOrder.astype(np.int32), rb.astype(np.float64),
neighbours.astype(np.int32), self.varYbar , y.astype(np.float64), matPl.astype(np.float64),
self.labels, np.array([varXh], dtype=np.float64),
self.currentValue,
self.nrlsSamples.astype(np.float64),
self.labelsSamples.astype(np.float64),
np.array([varXht], dtype=np.float64),
gTg.astype(np.float64),
beta.astype(np.float64), mean.astype(np.float64),
var.astype(np.float64), self.meanClassApost,
self.varClassApost, w.astype(np.int32), t1, t2,
cardClassCA.astype(np.int32),
self.nbClasses,
self.sampleLabelsFlag+0, self.iteration,
self.nbConditions,sampleWFlag)
self.countLabels(self.labels, self.voxIdx, self.cardClass)
def sub_sample_vol(image_file, dest_file, dsf, interpolation="continuous", verb_lvl=0):
from nipy.labs.datasets.volumes.volume_img import VolumeImg # ,CompositionError
pyhrf.verbose(verb_lvl, "Subsampling at dsf=%d, %s -> %s" % (dsf, image_file, dest_file))
interp = interpolation
data_src, src_meta = read_volume(image_file)
affine = src_meta[0]
daffine = dsf * affine
original_dtype = data_src.dtype
if np.issubdtype(np.int, data_src.dtype):
# to avoid error "array type 5 not supported" on int arrays ...
# data_src = np.asarray(np.asfarray(data_src), data_src.dtype)
data_src = np.asfarray(data_src)
if data_src.ndim == 4:
ref_vol = data_src[:, :, :, 0]
else:
ref_vol = data_src
# print 'ref_vol:', ref_vol.shape
img_src = VolumeImg(ref_vol, affine, "mine")
img_dest = img_src.as_volume_img(daffine, interpolation=interpolation)
# setup dest geometry:
dest_header = src_meta[1].copy()
# dest_header['sform'] = img_src.affine
dpixdim = np.array(src_meta[1]["pixdim"])
dpixdim[1:4] *= dsf
# print 'pixdim:', ','.join(map(str,src_meta[1]['pixdim']))
# print 'dpixdim:', ','.join(map(str,dpixdim))
dest_header["pixdim"] = list(dpixdim)
sh = ref_vol[::dsf, ::dsf, ::dsf, ...].shape
# print 'sh:', sh
dest_meta = (daffine, dest_header)
# do the resampling:
if data_src.ndim == 3:
vol = img_dest.get_data()[: sh[0], : sh[1], : sh[2]]
if ref_vol.dtype == np.int32 or np.allclose(np.round(ref_vol), ref_vol): # if input is int
vol = np.round(vol).astype(np.int32)
write_volume(vol, dest_file, dest_meta)
elif data_src.ndim == 4:
imgs = [VolumeImg(i, affine, "mine") for i in np.rollaxis(data_src, 3, 0)]
dvols = [i.as_volume_img(daffine, interpolation=interp).get_data() for i in imgs]
# print 'dvols[0]:', dvols[0].shape
dvols = np.array(dvols)
# print 'dvols:', dvols.shape
sub_vols = np.rollaxis(dvols[:, : sh[0], : sh[1], : sh[2]], 0, 4)
# print 'sub_vols:', sub_vols.shape
write_volume(sub_vols, dest_file, dest_meta)
else:
raise Exception("Nb of dims (%d) not handled. Only 3D or 4D" % data_src.ndim)
def manageMappingInit(self, shape, axes_names):
tans = self.dataInput.voxelMapping.getTargetAxesNames()
i = axes_names.index("voxel")
axes_names = axes_names[:i] + tans + axes_names[i + 1 :]
shape = shape[:i] + self.dataInput.finalShape + shape[i + 1 :]
pyhrf.verbose(5, "manageMappingInit returns :")
pyhrf.verbose(5, " -> sh: %s, axes_names: %s" % (str(shape), str(axes_names)))
return shape, axes_names
def to_ui_node(self, label, parent=None):
pyhrf.verbose(6, "Initable.to_ui_node(label=%s) ..." % label)
n = UiNode(label, parent, {"init_obj": self._init_obj, "type": "Initable"})
for pname, pval in self._init_parameters.iteritems():
pyhrf.verbose(6, "pname: %s, pval: %s" % (str(pname), str(pval)))
n.add_child(UiNode.from_py_object(pname, pval))
return n
def remote_mkdir(host, user, path):
import paramiko
pyhrf.verbose(1, 'Make remote dir %s@%s:%s ...' %(user,host,path))
ssh = paramiko.SSHClient()
known_hosts_file = os.path.join("~", ".ssh", "known_hosts")
ssh.load_host_keys(os.path.expanduser(known_hosts_file))
ssh.connect(host, username=user)
sftp = ssh.open_sftp()
sftp.mkdir(path)
