def test_max_sigmaH_prior(self):
D = 3
Thrf = 25.
dt = .5
TT, m_h = getCanoHRF(Thrf, dt)
m_h = m_h[:D]
m_H = np.array(m_h).astype(np.float64)
Sigma_H = np.ones((D, D), dtype=np.float64)
order = 2
D2 = vt.buildFiniteDiffMatrix(order, D)
R = np.dot(D2, D2) / pow(dt, 2*order)
gamma_h = 1000
sigmaH = vt.maximization_sigmaH_prior(D, Sigma_H, R, m_H, gamma_h)
def Main_vbjde_Extension_constrained_stable(graph, Y, Onsets, Thrf, K, TR, beta,
dt, scale=1, estimateSigmaH=True,
sigmaH=0.05, NitMax=-1,
NitMin=1, estimateBeta=True,
PLOT=False, contrasts=[],
computeContrast=False,
gamma_h=0):
""" Version modified by Lofti from Christine's version """
logger.info(
"Fast EM with C extension started ... Here is the stable version !")
np.random.seed(6537546)
# Initialize parameters
S = 100
if NitMax < 0:
NitMax = 100
gamma = 7.5 # 7.5
gradientStep = 0.003
MaxItGrad = 200
Thresh = 1e-5
# Initialize sizes vectors
D = np.int(np.ceil(Thrf / dt)) + 1
M = len(Onsets)
N = Y.shape[0]
J = Y.shape[1]
l = np.int(np.sqrt(J))
condition_names = []
# Neighbours
maxNeighbours = max([len(nl) for nl in graph])
neighboursIndexes = np.zeros((J, maxNeighbours), dtype=np.int32)
neighboursIndexes -= 1
for i in xrange(J):
neighboursIndexes[i, :len(graph[i])] = graph[i]
# Conditions
X = OrderedDict([])
for condition, Ons in Onsets.iteritems():
X[condition] = vt.compute_mat_X_2(N, TR, D, dt, Ons)
condition_names += [condition]
XX = np.zeros((M, N, D), dtype=np.int32)
nc = 0
for condition, Ons in Onsets.iteritems():
XX[nc, :, :] = X[condition]
nc += 1
# Covariance matrix
order = 2
D2 = vt.buildFiniteDiffMatrix(order, D)
R = np.dot(D2, D2) / pow(dt, 2 * order)
invR = np.linalg.inv(R)
Det_invR = np.linalg.det(invR)
Gamma = np.identity(N)
Det_Gamma = np.linalg.det(Gamma)
Crit_H = 1
Crit_Z = 1
Crit_A = 1
Crit_AH = 1
AH = np.zeros((J, M, D), dtype=np.float64)
AH1 = np.zeros((J, M, D), dtype=np.float64)
Crit_FreeEnergy = 1
cTime = []
cA = []
cH = []
cZ = []
cAH = []
CONTRAST = np.zeros((J, len(contrasts)), dtype=np.float64)
CONTRASTVAR = np.zeros((J, len(contrasts)), dtype=np.float64)
Q_barnCond = np.zeros((M, M, D, D), dtype=np.float64)
XGamma = np.zeros((M, D, N), dtype=np.float64)
m1 = 0
for k1 in X: # Loop over the M conditions
m2 = 0
for k2 in X:
Q_barnCond[m1, m2, :, :] = np.dot(
np.dot(X[k1].transpose(), Gamma), X[k2])
m2 += 1
XGamma[m1, :, :] = np.dot(X[k1].transpose(), Gamma)
m1 += 1
sigma_epsilone = np.ones(J)
logger.info(
"Labels are initialized by setting active probabilities to ones ...")
q_Z = np.zeros((M, K, J), dtype=np.float64)
q_Z[:, 1, :] = 1
q_Z1 = np.zeros((M, K, J), dtype=np.float64)
Z_tilde = q_Z.copy()
TT, m_h = getCanoHRF(Thrf, dt) # TODO: check
m_h = m_h[:D]
m_H = np.array(m_h).astype(np.float64)
m_H1 = np.array(m_h)
sigmaH1 = sigmaH
Sigma_H = np.ones((D, D), dtype=np.float64)
Beta = beta * np.ones((M), dtype=np.float64)
P = vt.PolyMat(N, 4, TR)
L = vt.polyFit(Y, TR, 4, P)
PL = np.dot(P, L)
y_tilde = Y - PL
Ndrift = L.shape[0]
sigma_M = np.ones((M, K), dtype=np.float64)
sigma_M[:, 0] = 0.5
sigma_M[:, 1] = 0.6
mu_M = np.zeros((M, K), dtype=np.float64)
for k in xrange(1, K):
mu_M[:, k] = 1 # InitMean
Sigma_A = np.zeros((M, M, J), np.float64)
for j in xrange(0, J):
Sigma_A[:, :, j] = 0.01 * np.identity(M)
m_A = np.zeros((J, M), dtype=np.float64)
m_A1 = np.zeros((J, M), dtype=np.float64)
for j in xrange(0, J):
for m in xrange(0, M):
for k in xrange(0, K):
m_A[j, m] += np.random.normal(mu_M[m, k],
np.sqrt(sigma_M[m, k])) * q_Z[m, k, j]
m_A1 = m_A
t1 = time.time()
##########################################################################
# VBJDE num. iter. minimum
ni = 0
while ((ni < NitMin + 1) or ((Crit_AH > Thresh) and (ni < NitMax))):
logger.info("------------------------------ Iteration n° " +
str(ni + 1) + " ------------------------------")
#####################
# EXPECTATION
#####################
# A
logger.info("E A step ...")
UtilsC.expectation_A(q_Z, mu_M, sigma_M, PL, sigma_epsilone, Gamma,
Sigma_H, Y, y_tilde, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N, K)
# crit. A
DIFF = np.reshape(m_A - m_A1, (M * J))
Crit_A = (
np.linalg.norm(DIFF) / np.linalg.norm(np.reshape(m_A1, (M * J)))) ** 2
cA += [Crit_A]
m_A1[:, :] = m_A[:, :]
# HRF h
UtilsC.expectation_H(XGamma, Q_barnCond, sigma_epsilone, Gamma, R, Sigma_H, Y,
y_tilde, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N, scale, sigmaH)
#m_H[0] = 0
#m_H[-1] = 0
# Constrain with optimization strategy
import cvxpy as cvx
m, n = Sigma_H.shape
Sigma_H_inv = np.linalg.inv(Sigma_H)
zeros_H = np.zeros_like(m_H[:, np.newaxis])
# Construct the problem. PRIMAL
h = cvx.Variable(n)
expression = cvx.quad_form(h - m_H[:, np.newaxis], Sigma_H_inv)
objective = cvx.Minimize(expression)
#constraints = [h[0] == 0, h[-1]==0, h >= zeros_H, cvx.square(cvx.norm(h,2))<=1]
constraints = [h[0] == 0, h[-1] == 0, cvx.square(cvx.norm(h, 2)) <= 1]
prob = cvx.Problem(objective, constraints)
result = prob.solve(verbose=0, solver=cvx.CVXOPT)
# Now we update the mean of h
m_H_old = m_H
Sigma_H_old = Sigma_H
m_H = np.squeeze(np.array((h.value)))
Sigma_H = np.zeros_like(Sigma_H)
# and the norm
h_norm += [np.linalg.norm(m_H)]
# crit. h
Crit_H = (np.linalg.norm(m_H - m_H1) / np.linalg.norm(m_H1)) ** 2
cH += [Crit_H]
m_H1[:] = m_H[:]
# crit. AH
for d in xrange(0, D):
AH[:, :, d] = m_A[:, :] * m_H[d]
DIFF = np.reshape(AH - AH1, (M * J * D))
Crit_AH = (np.linalg.norm(
DIFF) / (np.linalg.norm(np.reshape(AH1, (M * J * D))) + eps)) ** 2
cAH += [Crit_AH]
AH1[:, :, :] = AH[:, :, :]
# Z labels
logger.info("E Z step ...")
UtilsC.expectation_Z(Sigma_A, m_A, sigma_M, Beta, Z_tilde, mu_M,
q_Z, neighboursIndexes.astype(np.int32), M, J, K, maxNeighbours)
# crit. Z
DIFF = np.reshape(q_Z - q_Z1, (M * K * J))
Crit_Z = (np.linalg.norm(DIFF) /
(np.linalg.norm(np.reshape(q_Z1, (M * K * J))) + eps)) ** 2
cZ += [Crit_Z]
q_Z1[:, :, :] = q_Z[:, :, :]
#####################
# MAXIMIZATION
#####################
# HRF: Sigma_h
if estimateSigmaH:
logger.info("M sigma_H step ...")
if gamma_h > 0:
sigmaH = vt.maximization_sigmaH_prior(
D, Sigma_H, R, m_H, gamma_h)
else:
sigmaH = vt.maximization_sigmaH(D, Sigma_H, R, m_H)
logger.info('sigmaH = %s', str(sigmaH))
# (mu,sigma)
logger.info("M (mu,sigma) step ...")
mu_M, sigma_M = vt.maximization_mu_sigma(
mu_M, sigma_M, q_Z, m_A, K, M, Sigma_A)
# Drift L
UtilsC.maximization_L(
Y, m_A, m_H, L, P, XX.astype(np.int32), J, D, M, Ndrift, N)
PL = np.dot(P, L)
y_tilde = Y - PL
# Beta
if estimateBeta:
logger.info("estimating beta")
for m in xrange(0, M):
Beta[m] = UtilsC.maximization_beta(beta, q_Z[m, :, :].astype(np.float64), Z_tilde[m, :, :].astype(
np.float64), J, K, neighboursIndexes.astype(np.int32), gamma, maxNeighbours, MaxItGrad, gradientStep)
logger.info("End estimating beta")
logger.info(Beta)
# Sigma noise
logger.info("M sigma noise step ...")
UtilsC.maximization_sigma_noise(
Gamma, PL, sigma_epsilone, Sigma_H, Y, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N)
t02 = time.time()
cTime += [t02 - t1]
t2 = time.time()
##########################################################################
# PLOTS and SNR computation
if PLOT and 0:
font = {'size': 15}
matplotlib.rc('font', **font)
savefig('./HRF_Iter_CompMod.png')
hold(False)
figure(2)
plot(cAH[1:-1], 'lightblue')
hold(True)
plot(cFE[1:-1], 'm')
hold(False)
legend(('CAH', 'CFE'))
grid(True)
savefig('./Crit_CompMod.png')
figure(3)
plot(FreeEnergyArray)
grid(True)
savefig('./FreeEnergy_CompMod.png')
figure(4)
for m in xrange(M):
plot(SUM_q_Z_array[m])
hold(True)
hold(False)
savefig('./Sum_q_Z_Iter_CompMod.png')
figure(5)
for m in xrange(M):
plot(mu1_array[m])
hold(True)
hold(False)
savefig('./mu1_Iter_CompMod.png')
figure(6)
plot(h_norm_array)
savefig('./HRF_Norm_CompMod.png')
Data_save = xndarray(h_norm_array, ['Iteration'])
Data_save.save('./HRF_Norm_Comp.nii')
CompTime = t2 - t1
cTimeMean = CompTime / ni
"""
Norm = np.linalg.norm(m_H)
m_H /= Norm
Sigma_H /= Norm**2
sigmaH /= Norm**2
m_A *= Norm
Sigma_A *= Norm**2
mu_M *= Norm
sigma_M *= Norm**2
sigma_M = np.sqrt(np.sqrt(sigma_M))
"""
logger.info("Nb iterations to reach criterion: %d", ni)
logger.info("Computational time = %s min %s s", str(
np.int(CompTime // 60)), str(np.int(CompTime % 60)))
logger.info('mu_M: %f', mu_M)
logger.info('sigma_M: %f', sigma_M)
logger.info("sigma_H = %s", str(sigmaH))
logger.info("Beta = %s", str(Beta))
StimulusInducedSignal = vt.computeFit(m_H, m_A, X, J, N)
SNR = 20 * \
np.log(
np.linalg.norm(Y) / np.linalg.norm(Y - StimulusInducedSignal - PL))
SNR /= np.log(10.)
print 'SNR comp =', SNR
return ni, m_A, m_H, q_Z, sigma_epsilone, mu_M, sigma_M, Beta, L, PL, CONTRAST, CONTRASTVAR, cA[2:], cH[2:], cZ[2:], cAH[2:], cTime[2:], cTimeMean, Sigma_A, StimulusInducedSignal
def Main_vbjde_Extension_constrained(graph, Y, Onsets, Thrf, K, TR, beta,
dt, scale=1, estimateSigmaH=True,
sigmaH=0.05, NitMax=-1,
NitMin=1, estimateBeta=True,
PLOT=False, contrasts=[],
computeContrast=False,
gamma_h=0, estimateHRF=True,
TrueHrfFlag=False,
HrfFilename='hrf.nii',
estimateLabels=True,
LabelsFilename='labels.nii',
MFapprox=False, InitVar=0.5,
InitMean=2.0, MiniVEMFlag=False,
NbItMiniVem=5):
# VBJDE Function for BOLD with contraints
logger.info("Fast EM with C extension started ...")
np.random.seed(6537546)
##########################################################################
# INITIALIZATIONS
# Initialize parameters
tau1 = 0.0
tau2 = 0.0
S = 100
Init_sigmaH = sigmaH
Nb2Norm = 1
NormFlag = False
if NitMax < 0:
NitMax = 100
gamma = 7.5
#gamma_h = 1000
gradientStep = 0.003
MaxItGrad = 200
Thresh = 1e-5
Thresh_FreeEnergy = 1e-5
estimateLabels = True # WARNING!! They should be estimated
# Initialize sizes vectors
D = int(np.ceil(Thrf / dt)) + 1 # D = int(np.ceil(Thrf/dt))
M = len(Onsets)
N = Y.shape[0]
J = Y.shape[1]
l = int(np.sqrt(J))
condition_names = []
# Neighbours
maxNeighbours = max([len(nl) for nl in graph])
neighboursIndexes = np.zeros((J, maxNeighbours), dtype=np.int32)
neighboursIndexes -= 1
for i in xrange(J):
neighboursIndexes[i, :len(graph[i])] = graph[i]
# Conditions
X = OrderedDict([])
for condition, Ons in Onsets.iteritems():
X[condition] = vt.compute_mat_X_2(N, TR, D, dt, Ons)
condition_names += [condition]
XX = np.zeros((M, N, D), dtype=np.int32)
nc = 0
for condition, Ons in Onsets.iteritems():
XX[nc, :, :] = X[condition]
nc += 1
# Covariance matrix
order = 2
D2 = vt.buildFiniteDiffMatrix(order, D)
R = np.dot(D2, D2) / pow(dt, 2 * order)
invR = np.linalg.inv(R)
Det_invR = np.linalg.det(invR)
Gamma = np.identity(N)
Det_Gamma = np.linalg.det(Gamma)
p_Wtilde = np.zeros((M, K), dtype=np.float64)
p_Wtilde1 = np.zeros((M, K), dtype=np.float64)
p_Wtilde[:, 1] = 1
Crit_H = 1
Crit_Z = 1
Crit_A = 1
Crit_AH = 1
AH = np.zeros((J, M, D), dtype=np.float64)
AH1 = np.zeros((J, M, D), dtype=np.float64)
Crit_FreeEnergy = 1
cA = []
cH = []
cZ = []
cAH = []
FreeEnergy_Iter = []
cTime = []
cFE = []
SUM_q_Z = [[] for m in xrange(M)]
mu1 = [[] for m in xrange(M)]
h_norm = []
h_norm2 = []
CONTRAST = np.zeros((J, len(contrasts)), dtype=np.float64)
CONTRASTVAR = np.zeros((J, len(contrasts)), dtype=np.float64)
Q_barnCond = np.zeros((M, M, D, D), dtype=np.float64)
XGamma = np.zeros((M, D, N), dtype=np.float64)
m1 = 0
for k1 in X: # Loop over the M conditions
m2 = 0
for k2 in X:
Q_barnCond[m1, m2, :, :] = np.dot(
np.dot(X[k1].transpose(), Gamma), X[k2])
m2 += 1
XGamma[m1, :, :] = np.dot(X[k1].transpose(), Gamma)
m1 += 1
if MiniVEMFlag:
logger.info("MiniVEM to choose the best initialisation...")
"""InitVar, InitMean, gamma_h = MiniVEM_CompMod(Thrf,TR,dt,beta,Y,K,
gamma,gradientStep,
MaxItGrad,D,M,N,J,S,
maxNeighbours,
neighboursIndexes,
XX,X,R,Det_invR,Gamma,
Det_Gamma,
scale,Q_barnCond,XGamma,
NbItMiniVem,
sigmaH,estimateHRF)"""
InitVar, InitMean, gamma_h = vt.MiniVEM_CompMod(Thrf, TR, dt, beta, Y, K, gamma, gradientStep, MaxItGrad, D, M, N, J, S, maxNeighbours,
neighboursIndexes, XX, X, R, Det_invR, Gamma, Det_Gamma, p_Wtilde, scale, Q_barnCond, XGamma, tau1, tau2, NbItMiniVem, sigmaH, estimateHRF)
sigmaH = Init_sigmaH
sigma_epsilone = np.ones(J)
logger.info(
"Labels are initialized by setting active probabilities to ones ...")
q_Z = np.zeros((M, K, J), dtype=np.float64)
q_Z[:, 1, :] = 1
q_Z1 = np.zeros((M, K, J), dtype=np.float64)
Z_tilde = q_Z.copy()
# TT,m_h = getCanoHRF(Thrf-dt,dt) #TODO: check
TT, m_h = getCanoHRF(Thrf, dt) # TODO: check
m_h = m_h[:D]
m_H = np.array(m_h).astype(np.float64)
m_H1 = np.array(m_h)
sigmaH1 = sigmaH
if estimateHRF:
Sigma_H = np.ones((D, D), dtype=np.float64)
else:
Sigma_H = np.zeros((D, D), dtype=np.float64)
Beta = beta * np.ones((M), dtype=np.float64)
P = vt.PolyMat(N, 4, TR)
L = vt.polyFit(Y, TR, 4, P)
PL = np.dot(P, L)
y_tilde = Y - PL
Ndrift = L.shape[0]
sigma_M = np.ones((M, K), dtype=np.float64)
sigma_M[:, 0] = 0.5
sigma_M[:, 1] = 0.6
mu_M = np.zeros((M, K), dtype=np.float64)
for k in xrange(1, K):
mu_M[:, k] = InitMean
Sigma_A = np.zeros((M, M, J), np.float64)
for j in xrange(0, J):
Sigma_A[:, :, j] = 0.01 * np.identity(M)
m_A = np.zeros((J, M), dtype=np.float64)
m_A1 = np.zeros((J, M), dtype=np.float64)
for j in xrange(0, J):
for m in xrange(0, M):
for k in xrange(0, K):
m_A[j, m] += np.random.normal(mu_M[m, k],
np.sqrt(sigma_M[m, k])) * q_Z[m, k, j]
m_A1 = m_A
t1 = time.time()
##########################################################################
# VBJDE num. iter. minimum
ni = 0
while ((ni < NitMin) or (((Crit_FreeEnergy > Thresh_FreeEnergy) or (Crit_AH > Thresh)) and (ni < NitMax))):
logger.info("------------------------------ Iteration n° " +
str(ni + 1) + " ------------------------------")
#####################
# EXPECTATION
#####################
# A
logger.info("E A step ...")
UtilsC.expectation_A(q_Z, mu_M, sigma_M, PL, sigma_epsilone, Gamma,
Sigma_H, Y, y_tilde, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N, K)
val = np.reshape(m_A, (M * J))
val[np.where((val <= 1e-50) & (val > 0.0))] = 0.0
val[np.where((val >= -1e-50) & (val < 0.0))] = 0.0
# crit. A
DIFF = np.reshape(m_A - m_A1, (M * J))
# To avoid numerical problems
DIFF[np.where((DIFF < 1e-50) & (DIFF > 0.0))] = 0.0
# To avoid numerical problems
DIFF[np.where((DIFF > -1e-50) & (DIFF < 0.0))] = 0.0
Crit_A = (
np.linalg.norm(DIFF) / np.linalg.norm(np.reshape(m_A1, (M * J)))) ** 2
cA += [Crit_A]
m_A1[:, :] = m_A[:, :]
# HRF h
if estimateHRF:
################################
# HRF ESTIMATION
################################
UtilsC.expectation_H(XGamma, Q_barnCond, sigma_epsilone, Gamma, R, Sigma_H, Y,
y_tilde, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N, scale, sigmaH)
import cvxpy as cvx
m, n = Sigma_H.shape
Sigma_H_inv = np.linalg.inv(Sigma_H)
zeros_H = np.zeros_like(m_H[:, np.newaxis])
# Construct the problem. PRIMAL
h = cvx.Variable(n)
expression = cvx.quad_form(h - m_H[:, np.newaxis], Sigma_H_inv)
objective = cvx.Minimize(expression)
#constraints = [h[0] == 0, h[-1]==0, h >= zeros_H, cvx.square(cvx.norm(h,2))<=1]
constraints = [
h[0] == 0, h[-1] == 0, cvx.square(cvx.norm(h, 2)) <= 1]
prob = cvx.Problem(objective, constraints)
result = prob.solve(verbose=0, solver=cvx.CVXOPT)
# Now we update the mean of h
m_H_old = m_H
Sigma_H_old = Sigma_H
m_H = np.squeeze(np.array((h.value)))
Sigma_H = np.zeros_like(Sigma_H)
h_norm += [np.linalg.norm(m_H)]
# print 'h_norm = ', h_norm
# Plotting HRF
if PLOT and ni >= 0:
import matplotlib.pyplot as plt
plt.figure(M + 1)
plt.plot(m_H)
plt.hold(True)
else:
if TrueHrfFlag:
#TrueVal, head = read_volume(HrfFilename)
TrueVal, head = read_volume(HrfFilename)[:, 0, 0, 0]
print TrueVal
print TrueVal.shape
m_H = TrueVal
# crit. h
Crit_H = (np.linalg.norm(m_H - m_H1) / np.linalg.norm(m_H1)) ** 2
cH += [Crit_H]
m_H1[:] = m_H[:]
# crit. AH
for d in xrange(0, D):
AH[:, :, d] = m_A[:, :] * m_H[d]
DIFF = np.reshape(AH - AH1, (M * J * D))
# To avoid numerical problems
DIFF[np.where((DIFF < 1e-50) & (DIFF > 0.0))] = 0.0
# To avoid numerical problems
DIFF[np.where((DIFF > -1e-50) & (DIFF < 0.0))] = 0.0
if np.linalg.norm(np.reshape(AH1, (M * J * D))) == 0:
Crit_AH = 1000000000.
else:
Crit_AH = (
np.linalg.norm(DIFF) / np.linalg.norm(np.reshape(AH1, (M * J * D)))) ** 2
cAH += [Crit_AH]
AH1[:, :, :] = AH[:, :, :]
# Z labels
if estimateLabels:
logger.info("E Z step ...")
# WARNING!!! ParsiMod gives better results, but we need the other
# one.
if MFapprox:
UtilsC.expectation_Z(Sigma_A, m_A, sigma_M, Beta, Z_tilde, mu_M, q_Z, neighboursIndexes.astype(
np.int32), M, J, K, maxNeighbours)
if not MFapprox:
UtilsC.expectation_Z_ParsiMod_RVM_and_CompMod(
Sigma_A, m_A, sigma_M, Beta, mu_M, q_Z, neighboursIndexes.astype(np.int32), M, J, K, maxNeighbours)
else:
logger.info("Using True Z ...")
TrueZ = read_volume(LabelsFilename)
for m in xrange(M):
q_Z[m, 1, :] = np.reshape(TrueZ[0][:, :, :, m], J)
q_Z[m, 0, :] = 1 - q_Z[m, 1, :]
# crit. Z
val = np.reshape(q_Z, (M * K * J))
val[np.where((val <= 1e-50) & (val > 0.0))] = 0.0
DIFF = np.reshape(q_Z - q_Z1, (M * K * J))
# To avoid numerical problems
DIFF[np.where((DIFF < 1e-50) & (DIFF > 0.0))] = 0.0
# To avoid numerical problems
DIFF[np.where((DIFF > -1e-50) & (DIFF < 0.0))] = 0.0
if np.linalg.norm(np.reshape(q_Z1, (M * K * J))) == 0:
Crit_Z = 1000000000.
else:
Crit_Z = (
np.linalg.norm(DIFF) / np.linalg.norm(np.reshape(q_Z1, (M * K * J)))) ** 2
cZ += [Crit_Z]
q_Z1 = q_Z
#####################
# MAXIMIZATION
#####################
# HRF: Sigma_h
if estimateHRF:
if estimateSigmaH:
logger.info("M sigma_H step ...")
if gamma_h > 0:
sigmaH = vt.maximization_sigmaH_prior(
D, Sigma_H_old, R, m_H_old, gamma_h)
else:
sigmaH = vt.maximization_sigmaH(D, Sigma_H, R, m_H)
logger.info('sigmaH = %s', str(sigmaH))
# (mu,sigma)
logger.info("M (mu,sigma) step ...")
mu_M, sigma_M = vt.maximization_mu_sigma(
mu_M, sigma_M, q_Z, m_A, K, M, Sigma_A)
for m in xrange(M):
SUM_q_Z[m] += [sum(q_Z[m, 1, :])]
mu1[m] += [mu_M[m, 1]]
# Drift L
UtilsC.maximization_L(
Y, m_A, m_H, L, P, XX.astype(np.int32), J, D, M, Ndrift, N)
PL = np.dot(P, L)
y_tilde = Y - PL
# Beta
if estimateBeta:
logger.info("estimating beta")
for m in xrange(0, M):
if MFapprox:
Beta[m] = UtilsC.maximization_beta(beta, q_Z[m, :, :].astype(np.float64), Z_tilde[m, :, :].astype(
np.float64), J, K, neighboursIndexes.astype(np.int32), gamma, maxNeighbours, MaxItGrad, gradientStep)
if not MFapprox:
#Beta[m] = UtilsC.maximization_beta(beta,q_Z[m,:,:].astype(np.float64),q_Z[m,:,:].astype(np.float64),J,K,neighboursIndexes.astype(int32),gamma,maxNeighbours,MaxItGrad,gradientStep)
Beta[m] = UtilsC.maximization_beta_CB(beta, q_Z[m, :, :].astype(
np.float64), J, K, neighboursIndexes.astype(np.int32), gamma, maxNeighbours, MaxItGrad, gradientStep)
logger.info("End estimating beta")
logger.info(Beta)
# Sigma noise
logger.info("M sigma noise step ...")
UtilsC.maximization_sigma_noise(
Gamma, PL, sigma_epsilone, Sigma_H, Y, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N)
#### Computing Free Energy ####
if ni > 0:
FreeEnergy1 = FreeEnergy
"""FreeEnergy = vt.Compute_FreeEnergy(y_tilde,m_A,Sigma_A,mu_M,sigma_M,
m_H,Sigma_H,R,Det_invR,sigmaH,
p_Wtilde,q_Z,neighboursIndexes,
maxNeighbours,Beta,sigma_epsilone,
XX,Gamma,Det_Gamma,XGamma,J,D,M,
N,K,S,"CompMod")"""
FreeEnergy = vt.Compute_FreeEnergy(y_tilde, m_A, Sigma_A, mu_M, sigma_M, m_H, Sigma_H, R, Det_invR, sigmaH, p_Wtilde, tau1,
tau2, q_Z, neighboursIndexes, maxNeighbours, Beta, sigma_epsilone, XX, Gamma, Det_Gamma, XGamma, J, D, M, N, K, S, "CompMod")
if ni > 0:
Crit_FreeEnergy = (FreeEnergy1 - FreeEnergy) / FreeEnergy1
FreeEnergy_Iter += [FreeEnergy]
cFE += [Crit_FreeEnergy]
# Update index
ni += 1
t02 = time.time()
cTime += [t02 - t1]
t2 = time.time()
##########################################################################
# PLOTS and SNR computation
FreeEnergyArray = np.zeros((ni), dtype=np.float64)
for i in xrange(ni):
FreeEnergyArray[i] = FreeEnergy_Iter[i]
SUM_q_Z_array = np.zeros((M, ni), dtype=np.float64)
mu1_array = np.zeros((M, ni), dtype=np.float64)
h_norm_array = np.zeros((ni), dtype=np.float64)
for m in xrange(M):
for i in xrange(ni):
SUM_q_Z_array[m, i] = SUM_q_Z[m][i]
mu1_array[m, i] = mu1[m][i]
h_norm_array[i] = h_norm[i]
if PLOT and 0:
import matplotlib.pyplot as plt
import matplotlib
font = {'size': 15}
matplotlib.rc('font', **font)
plt.savefig('./HRF_Iter_CompMod.png')
plt.hold(False)
plt.figure(2)
plt.plot(cAH[1:-1], 'lightblue')
plt.hold(True)
plt.plot(cFE[1:-1], 'm')
plt.hold(False)
#plt.legend( ('CA','CH', 'CZ', 'CAH', 'CFE') )
plt.legend(('CAH', 'CFE'))
plt.grid(True)
plt.savefig('./Crit_CompMod.png')
plt.figure(3)
plt.plot(FreeEnergyArray)
plt.grid(True)
plt.savefig('./FreeEnergy_CompMod.png')
plt.figure(4)
for m in xrange(M):
plt.plot(SUM_q_Z_array[m])
plt.hold(True)
plt.hold(False)
#plt.legend( ('m=0','m=1', 'm=2', 'm=3') )
#plt.legend( ('m=0','m=1') )
plt.savefig('./Sum_q_Z_Iter_CompMod.png')
plt.figure(5)
for m in xrange(M):
plt.plot(mu1_array[m])
plt.hold(True)
plt.hold(False)
plt.savefig('./mu1_Iter_CompMod.png')
plt.figure(6)
plt.plot(h_norm_array)
plt.savefig('./HRF_Norm_CompMod.png')
Data_save = xndarray(h_norm_array, ['Iteration'])
Data_save.save('./HRF_Norm_Comp.nii')
CompTime = t2 - t1
cTimeMean = CompTime / ni
sigma_M = np.sqrt(np.sqrt(sigma_M))
logger.info("Nb iterations to reach criterion: %d", ni)
logger.info("Computational time = %s min %s s", str(
int(CompTime // 60)), str(int(CompTime % 60)))
# print "Computational time = " + str(int( CompTime//60 ) ) + " min " + str(int(CompTime%60)) + " s"
# print "sigma_H = " + str(sigmaH)
logger.info('mu_M: %f', mu_M)
logger.info('sigma_M: %f', sigma_M)
logger.info("sigma_H = %s" + str(sigmaH))
logger.info("Beta = %s" + str(Beta))
StimulusInducedSignal = vt.computeFit(m_H, m_A, X, J, N)
SNR = 20 * \
np.log(
np.linalg.norm(Y) / np.linalg.norm(Y - StimulusInducedSignal - PL))
SNR /= np.log(10.)
logger.info("SNR = %d", SNR)
return ni, m_A, m_H, q_Z, sigma_epsilone, mu_M, sigma_M, Beta, L, PL, CONTRAST, CONTRASTVAR, cA[2:], cH[2:], cZ[2:], cAH[2:], cTime[2:], cTimeMean, Sigma_A, StimulusInducedSignal, FreeEnergyArray
def jde_vem_bold(graph, bold_data, onsets, durations, hrf_duration, nb_classes,
tr, beta, dt, estimate_sigma_h=True, sigma_h=0.05,
it_max=-1, it_min=0, estimate_beta=True, contrasts=None,
compute_contrasts=False, hrf_hyperprior=0, estimate_hrf=True,
constrained=False, zero_constraint=True, drifts_type="poly",
seed=6537546):
"""This is the main function that computes the VEM analysis on BOLD data.
This function uses optimized python functions.
Parameters
----------
graph : ndarray of lists
represents the neighbours indexes of each voxels index
bold_data : ndarray, shape (nb_scans, nb_voxels)
raw data
onsets : dict
dictionnary of onsets
durations : # TODO
# TODO
hrf_duration : float
hrf total time duration (in s)
nb_classes : int
the number of classes to classify the nrls. This parameter is provided
for development purposes as most of the algorithm implies two classes
tr : float
time of repetition
beta : float
the initial value of beta
dt : float
hrf temporal precision
estimate_sigma_h : bool, optional
toggle estimation of sigma H
sigma_h : float, optional
initial or fixed value of sigma H
it_max : int, optional
maximal computed iteration number
it_min : int, optional
minimal computed iteration number
estimate_beta : bool, optional
toggle the estimation of beta
contrasts : OrderedDict, optional
dict of contrasts to compute
compute_contrasts : bool, optional
if True, compute the contrasts defined in contrasts
hrf_hyperprior : float
# TODO
estimate_hrf : bool, optional
if True, estimate the HRF for each parcel, if False use the canonical HRF
constrained : bool, optional
if True, add a constrains the l2 norm of the HRF to 1
drifts_type : str, optional
set the drifts basis type used. Can be "poly" for polynomial or "cos"
for cosine
seed : int, optional
seed used by numpy to initialize random generator number
Returns
-------
loop : int
number of iterations before convergence
nrls_mean : ndarray, shape (nb_voxels, nb_conditions)
Neural response level mean value
hrf_mean : ndarray, shape (hrf_len,)
Hemodynamic response function mean value
hrf_covar : ndarray, shape (hrf_len, hrf_len)
Covariance matrix of the HRF
labels_proba : ndarray, shape (nb_conditions, nb_classes, nb_voxels)
probability of voxels being in one class
noise_var : ndarray, shape (nb_voxels,)
estimated noise variance
nrls_class_mean : ndarray, shape (nb_conditions, nb_classes)
estimated mean value of the gaussians of the classes
nrls_class_var : ndarray, shape (nb_conditions, nb_classes)
estimated variance of the gaussians of the classes
beta : ndarray, shape (nb_conditions,)
estimated beta
drift_coeffs : ndarray, shape (# TODO)
estimated coefficient of the drifts
drift : ndarray, shape (# TODO)
estimated drifts
contrasts_mean : ndarray, shape (nb_voxels, len(contrasts))
Contrasts computed from NRLs
contrasts_var : ndarray, shape (nb_voxels, len(contrasts))
Variance of the contrasts
compute_time : list
computation time of each iteration
compute_time_mean : float
computation mean time over iterations
nrls_covar : ndarray, shape (nb_conditions, nb_conditions, nb_voxels)
# TODO
stimulus_induced_signal : ndarray, shape (nb_scans, nb_voxels)
# TODO
mahalanobis_zero : float
Mahalanobis distance between estimated hrf_mean and the null vector
mahalanobis_cano : float
Mahalanobis distance between estimated hrf_mean and the canonical HRF
mahalanobis_diff : float
difference between mahalanobis_cano and mahalanobis_diff
mahalanobis_prod : float
product of mahalanobis_cano and mahalanobis_diff
ppm_a_nrl : ndarray, shape (nb_voxels,)
The posterior probability map using an alpha
ppm_g_nrl : ndarray, shape (nb_voxels,)
# TODO
ppm_a_contrasts : ndarray, shape (nb_voxels,)
# TODO
ppm_g_contrasts : ndarray, shape (nb_voxels,)
# TODO
variation_coeff : float
coefficient of variation of the HRF
free_energy : list
# TODO
Notes
-----
See `A novel definition of the multivariate coefficient of variation
<http://onlinelibrary.wiley.com/doi/10.1002/bimj.201000030/abstract>`_
article for more information about the coefficient of variation.
"""
logger.info("VEM started.")
if not contrasts:
contrasts = OrderedDict()
np.random.seed(seed)
nb_2_norm = 1
normalizing = False
regularizing = False
if it_max <= 0:
it_max = 100
gamma = 7.5
thresh_free_energy = 1e-4
# Initialize sizes vectors
hrf_len = np.int(np.ceil(hrf_duration / dt)) + 1
nb_conditions = len(onsets)
nb_scans = bold_data.shape[0]
nb_voxels = bold_data.shape[1]
X, occurence_matrix, condition_names = vt.create_conditions(
onsets, durations, nb_conditions, nb_scans, hrf_len, tr, dt
)
neighbours_indexes = vt.create_neighbours(graph)
order = 2
if regularizing:
regularization = np.ones(hrf_len)
regularization[hrf_len//3:hrf_len//2] = 2
regularization[hrf_len//2:2*hrf_len//3] = 5
regularization[2*hrf_len//3:3*hrf_len//4] = 7
regularization[3*hrf_len//4:] = 10
# regularization[hrf_len//2:] = 10
else:
regularization = None
d2 = vt.buildFiniteDiffMatrix(order, hrf_len, regularization)
hrf_regu_prior_inv = d2.T.dot(d2) / pow(dt, 2 * order)
if estimate_hrf and zero_constraint:
hrf_len = hrf_len - 2
hrf_regu_prior_inv = hrf_regu_prior_inv[1:-1, 1:-1]
occurence_matrix = occurence_matrix[:, :, 1:-1]
noise_struct = np.identity(nb_scans)
free_energy = [1.]
free_energy_crit = [1.]
compute_time = []
noise_var = np.ones(nb_voxels)
labels_proba = np.zeros((nb_conditions, nb_classes, nb_voxels), dtype=np.float64)
logger.info("Labels are initialized by setting everything to {}".format(1./nb_classes))
labels_proba[:, :, :] = 1./nb_classes
m_h = getCanoHRF(hrf_duration, dt)[1][:hrf_len]
hrf_mean = np.array(m_h).astype(np.float64)
if estimate_hrf:
hrf_covar = np.identity(hrf_len, dtype=np.float64)
else:
hrf_covar = np.zeros((hrf_len, hrf_len), dtype=np.float64)
beta = beta * np.ones((nb_conditions), dtype=np.float64)
beta_list = []
beta_list.append(beta.copy())
if drifts_type == "poly":
drift_basis = vt.poly_drifts_basis(nb_scans, 4, tr)
elif drifts_type == "cos":
drift_basis = vt.cosine_drifts_basis(nb_scans, 64, tr)
drift_coeffs = vt.drifts_coeffs_fit(bold_data, drift_basis)
drift = drift_basis.dot(drift_coeffs)
bold_data_drift = bold_data - drift
# Parameters Gaussian mixtures
nrls_class_mean = 2 * np.ones((nb_conditions, nb_classes))
nrls_class_mean[:, 0] = 0
nrls_class_var = 0.3 * np.ones((nb_conditions, nb_classes), dtype=np.float64)
nrls_mean = (np.random.normal(
nrls_class_mean, nrls_class_var)[:, :, np.newaxis] * labels_proba).sum(axis=1).T
nrls_covar = (np.identity(nb_conditions)[:, :, np.newaxis] + np.zeros((1, 1, nb_voxels)))
start_time = time.time()
loop = 0
while (loop <= it_min or
((np.asarray(free_energy_crit[-5:]) > thresh_free_energy).any()
and loop < it_max)):
logger.info("{:-^80}".format(" Iteration n°"+str(loop+1)+" "))
logger.info("Expectation A step...")
logger.debug("Before: nrls_mean = %s, nrls_covar = %s", nrls_mean, nrls_covar)
nrls_mean, nrls_covar = vt.nrls_expectation(
hrf_mean, nrls_mean, occurence_matrix, noise_struct, labels_proba,
nrls_class_mean, nrls_class_var, nb_conditions, bold_data_drift, nrls_covar,
hrf_covar, noise_var)
logger.debug("After: nrls_mean = %s, nrls_covar = %s", nrls_mean, nrls_covar)
logger.info("Expectation Z step...")
logger.debug("Before: labels_proba = %s, labels_proba = %s", labels_proba, labels_proba)
labels_proba = vt.labels_expectation(
nrls_covar, nrls_mean, nrls_class_var, nrls_class_mean, beta,
labels_proba, neighbours_indexes, nb_conditions, nb_classes,
nb_voxels, parallel=True)
logger.debug("After: labels_proba = %s, labels_proba = %s", labels_proba, labels_proba)
if estimate_hrf:
logger.info("Expectation H step...")
logger.debug("Before: hrf_mean = %s, hrf_covar = %s", hrf_mean, hrf_covar)
hrf_mean, hrf_covar = vt.hrf_expectation(
nrls_covar, nrls_mean, occurence_matrix, noise_struct,
hrf_regu_prior_inv, sigma_h, nb_voxels, bold_data_drift, noise_var)
if constrained:
hrf_mean = vt.norm1_constraint(hrf_mean, hrf_covar)
hrf_covar[:] = 0
logger.debug("After: hrf_mean = %s, hrf_covar = %s", hrf_mean, hrf_covar)
# Normalizing H at each nb_2_norm iterations:
if not constrained and normalizing:
# Normalizing is done before sigma_h, nrls_class_mean and nrls_class_var estimation
# we should not include them in the normalisation step
if (loop + 1) % nb_2_norm == 0:
hrf_norm = np.linalg.norm(hrf_mean)
hrf_mean /= hrf_norm
hrf_covar /= hrf_norm ** 2
nrls_mean *= hrf_norm
nrls_covar *= hrf_norm ** 2
if estimate_hrf and estimate_sigma_h:
logger.info("Maximization sigma_H step...")
logger.debug("Before: sigma_h = %s", sigma_h)
if hrf_hyperprior > 0:
sigma_h = vt.maximization_sigmaH_prior(hrf_len, hrf_covar,
hrf_regu_prior_inv,
hrf_mean, hrf_hyperprior)
else:
sigma_h = vt.maximization_sigmaH(hrf_len, hrf_covar,
hrf_regu_prior_inv, hrf_mean)
logger.debug("After: sigma_h = %s", sigma_h)
logger.info("Maximization (mu,sigma) step...")
logger.debug("Before: nrls_class_mean = %s, nrls_class_var = %s",
nrls_class_mean, nrls_class_var)
nrls_class_mean, nrls_class_var = vt.maximization_class_proba(
labels_proba, nrls_mean, nrls_covar
)
logger.debug("After: nrls_class_mean = %s, nrls_class_var = %s",
nrls_class_mean, nrls_class_var)
logger.info("Maximization L step...")
logger.debug("Before: drift_coeffs = %s", drift_coeffs)
drift_coeffs = vt.maximization_drift_coeffs(
bold_data, nrls_mean, occurence_matrix, hrf_mean, noise_struct, drift_basis
)
logger.debug("After: drift_coeffs = %s", drift_coeffs)
drift = drift_basis.dot(drift_coeffs)
bold_data_drift = bold_data - drift
if estimate_beta:
logger.info("Maximization beta step...")
for cond_nb in xrange(0, nb_conditions):
beta[cond_nb], success = vt.beta_maximization(
beta[cond_nb]*np.ones((1,)), labels_proba[cond_nb, :, :],
neighbours_indexes, gamma
)
beta_list.append(beta.copy())
logger.debug("beta = %s", str(beta))
logger.info("Maximization sigma noise step...")
noise_var = vt.maximization_noise_var(
occurence_matrix, hrf_mean, hrf_covar, nrls_mean, nrls_covar,
noise_struct, bold_data_drift, nb_scans
)
#### Computing Free Energy ####
free_energy.append(vt.free_energy_computation(
nrls_mean, nrls_covar, hrf_mean, hrf_covar, hrf_len, labels_proba,
bold_data_drift, occurence_matrix, noise_var, noise_struct, nb_conditions,
nb_voxels, nb_scans, nb_classes, nrls_class_mean, nrls_class_var, neighbours_indexes,
beta, sigma_h, np.linalg.inv(hrf_regu_prior_inv), hrf_regu_prior_inv, gamma, hrf_hyperprior
))
free_energy_crit.append(abs((free_energy[-2] - free_energy[-1]) /
free_energy[-2]))
logger.info("Convergence criteria: %f (Threshold = %f)",
free_energy_crit[-1], thresh_free_energy)
loop += 1
compute_time.append(time.time() - start_time)
compute_time_mean = compute_time[-1] / loop
mahalanobis_zero = np.nan
mahalanobis_cano = np.nan
mahalanobis_diff = np.nan
mahalanobis_prod = np.nan
variation_coeff = np.nan
if estimate_hrf and not constrained and not normalizing:
hrf_norm = np.linalg.norm(hrf_mean)
hrf_mean /= hrf_norm
hrf_covar /= hrf_norm ** 2
sigma_h /= hrf_norm ** 2
nrls_mean *= hrf_norm
nrls_covar *= hrf_norm ** 2
nrls_class_mean *= hrf_norm
nrls_class_var *= hrf_norm ** 2
mahalanobis_zero = mahalanobis(hrf_mean, np.zeros_like(hrf_mean),
np.linalg.inv(hrf_covar))
mahalanobis_cano = mahalanobis(hrf_mean, m_h, np.linalg.inv(hrf_covar))
mahalanobis_diff = mahalanobis_cano - mahalanobis_zero
mahalanobis_prod = mahalanobis_cano * mahalanobis_zero
variation_coeff = np.sqrt((hrf_mean.T.dot(hrf_covar).dot(hrf_mean))
/(hrf_mean.T.dot(hrf_mean))**2)
if estimate_hrf and zero_constraint:
hrf_mean = np.concatenate(([0], hrf_mean, [0]))
# when using the zero constraint the hrf covariance is fill with
# arbitrary zeros around the matrix, this is maybe a bad idea if we need
# it for later computation...
hrf_covar = np.concatenate(
(np.zeros((hrf_covar.shape[0], 1)), hrf_covar, np.zeros((hrf_covar.shape[0], 1))),
axis=1
)
hrf_covar = np.concatenate(
(np.zeros((1, hrf_covar.shape[1])), hrf_covar, np.zeros((1, hrf_covar.shape[1]))),
axis=0
)
if estimate_hrf:
(delay_of_response, delay_of_undershoot, dispersion_of_response,
dispersion_of_undershoot, ratio_resp_under, delay) = vt.fit_hrf_two_gammas(
hrf_mean, dt, hrf_duration
)
else:
(delay_of_response, delay_of_undershoot, dispersion_of_response,
dispersion_of_undershoot, ratio_resp_under, delay) = (None, None, None,
None, None, None)
ppm_a_nrl, ppm_g_nrl = vt.ppms_computation(
nrls_mean, np.diagonal(nrls_covar), nrls_class_mean, nrls_class_var,
threshold_a="intersect"
)
#+++++++++++++++++++++++ calculate contrast maps and variance +++++++++++++++++++++++#
nb_contrasts = len(contrasts)
if compute_contrasts and nb_contrasts > 0:
logger.info('Computing contrasts ...')
(contrasts_mean,
contrasts_var,
contrasts_class_mean,
contrasts_class_var) = vt.contrasts_mean_var_classes(
contrasts, condition_names, nrls_mean, nrls_covar,
nrls_class_mean, nrls_class_var, nb_contrasts, nb_classes, nb_voxels
)
ppm_a_contrasts, ppm_g_contrasts = vt.ppms_computation(
contrasts_mean, contrasts_var, contrasts_class_mean, contrasts_class_var
)
logger.info('Done computing contrasts.')
else:
(contrasts_mean, contrasts_var, contrasts_class_mean,
contrasts_class_var, ppm_a_contrasts, ppm_g_contrasts) = (None, None,
None, None,
None, None)
#+++++++++++++++++++++++ calculate contrast maps and variance +++++++++++++++++++++++#
logger.info("Nb iterations to reach criterion: %d", loop)
logger.info("Computational time = %s min %s s",
*(str(int(x)) for x in divmod(compute_time[-1], 60)))
logger.debug('nrls_class_mean: %s', nrls_class_mean)
logger.debug('nrls_class_var: %s', nrls_class_var)
logger.debug("sigma_H = %s", str(sigma_h))
logger.debug("beta = %s", str(beta))
stimulus_induced_signal = vt.computeFit(hrf_mean, nrls_mean, X, nb_voxels, nb_scans)
snr = 20 * np.log(
np.linalg.norm(bold_data.astype(np.float))
/ np.linalg.norm((bold_data - stimulus_induced_signal - drift).astype(np.float))
)
snr /= np.log(10.)
logger.info('snr comp = %f', snr)
# ,FreeEnergyArray
return (loop, nrls_mean, hrf_mean, hrf_covar, labels_proba, noise_var,
nrls_class_mean, nrls_class_var, beta, drift_coeffs, drift,
contrasts_mean, contrasts_var, compute_time[2:], compute_time_mean,
nrls_covar, stimulus_induced_signal, mahalanobis_zero,
mahalanobis_cano, mahalanobis_diff, mahalanobis_prod, ppm_a_nrl,
ppm_g_nrl, ppm_a_contrasts, ppm_g_contrasts, variation_coeff,
free_energy[1:], free_energy_crit[1:], beta_list[1:],
delay_of_response, delay_of_undershoot, dispersion_of_response,
dispersion_of_undershoot, ratio_resp_under, delay)
def jde_vem_bold(graph,
bold_data,
onsets,
durations,
hrf_duration,
nb_classes,
tr,
beta,
dt,
estimate_sigma_h=True,
sigma_h=0.05,
it_max=-1,
it_min=0,
estimate_beta=True,
contrasts=None,
compute_contrasts=False,
hrf_hyperprior=0,
estimate_hrf=True,
constrained=False,
zero_constraint=True,
drifts_type="poly",
seed=6537546):
"""This is the main function that computes the VEM analysis on BOLD data.
This function uses optimized python functions.
Parameters
----------
graph : ndarray of lists
represents the neighbours indexes of each voxels index
bold_data : ndarray, shape (nb_scans, nb_voxels)
raw data
onsets : dict
dictionnary of onsets
durations : # TODO
# TODO
hrf_duration : float
hrf total time duration (in s)
nb_classes : int
the number of classes to classify the nrls. This parameter is provided
for development purposes as most of the algorithm implies two classes
tr : float
time of repetition
beta : float
the initial value of beta
dt : float
hrf temporal precision
estimate_sigma_h : bool, optional
toggle estimation of sigma H
sigma_h : float, optional
initial or fixed value of sigma H
it_max : int, optional
maximal computed iteration number
it_min : int, optional
minimal computed iteration number
estimate_beta : bool, optional
toggle the estimation of beta
contrasts : OrderedDict, optional
dict of contrasts to compute
compute_contrasts : bool, optional
if True, compute the contrasts defined in contrasts
hrf_hyperprior : float
# TODO
estimate_hrf : bool, optional
if True, estimate the HRF for each parcel, if False use the canonical HRF
constrained : bool, optional
if True, add a constrains the l2 norm of the HRF to 1
zero_constraint : bool, optional
if True, add zeros to the beginning and the end of the estimated HRF.
drifts_type : str, optional
set the drifts basis type used. Can be "poly" for polynomial or "cos"
for cosine
seed : int, optional
seed used by numpy to initialize random generator number
Returns
-------
loop : int
number of iterations before convergence
nrls_mean : ndarray, shape (nb_voxels, nb_conditions)
Neural response level mean value
hrf_mean : ndarray, shape (hrf_len,)
Hemodynamic response function mean value
hrf_covar : ndarray, shape (hrf_len, hrf_len)
Covariance matrix of the HRF
labels_proba : ndarray, shape (nb_conditions, nb_classes, nb_voxels)
probability of voxels being in one class
noise_var : ndarray, shape (nb_voxels,)
estimated noise variance
nrls_class_mean : ndarray, shape (nb_conditions, nb_classes)
estimated mean value of the gaussians of the classes
nrls_class_var : ndarray, shape (nb_conditions, nb_classes)
estimated variance of the gaussians of the classes
beta : ndarray, shape (nb_conditions,)
estimated beta
drift_coeffs : ndarray, shape (# TODO)
estimated coefficient of the drifts
drift : ndarray, shape (# TODO)
estimated drifts
contrasts_mean : ndarray, shape (nb_voxels, len(contrasts))
Contrasts computed from NRLs
contrasts_var : ndarray, shape (nb_voxels, len(contrasts))
Variance of the contrasts
compute_time : list
computation time of each iteration
compute_time_mean : float
computation mean time over iterations
nrls_covar : ndarray, shape (nb_conditions, nb_conditions, nb_voxels)
# TODO
stimulus_induced_signal : ndarray, shape (nb_scans, nb_voxels)
# TODO
mahalanobis_zero : float
Mahalanobis distance between estimated hrf_mean and the null vector
mahalanobis_cano : float
Mahalanobis distance between estimated hrf_mean and the canonical HRF
mahalanobis_diff : float
difference between mahalanobis_cano and mahalanobis_diff
mahalanobis_prod : float
product of mahalanobis_cano and mahalanobis_diff
ppm_a_nrl : ndarray, shape (nb_voxels,)
The posterior probability map using an alpha
ppm_g_nrl : ndarray, shape (nb_voxels,)
# TODO
ppm_a_contrasts : ndarray, shape (nb_voxels,)
# TODO
ppm_g_contrasts : ndarray, shape (nb_voxels,)
# TODO
variation_coeff : float
coefficient of variation of the HRF
free_energy : list
# TODO
Notes
-----
See `A novel definition of the multivariate coefficient of variation
<http://onlinelibrary.wiley.com/doi/10.1002/bimj.201000030/abstract>`_
article for more information about the coefficient of variation.
"""
logger.info("VEM started.")
if not contrasts:
contrasts = OrderedDict()
np.random.seed(seed)
nb_2_norm = 1
normalizing = False
regularizing = False
if it_max <= 0:
it_max = 100
gamma = 7.5
# Initialize sizes vectors
hrf_len = np.int(np.ceil(hrf_duration / dt)) + 1
nb_conditions = len(onsets)
nb_scans = bold_data.shape[0]
nb_voxels = bold_data.shape[1]
X, occurence_matrix, condition_names = vt.create_conditions(
onsets, durations, nb_conditions, nb_scans, hrf_len, tr, dt)
neighbours_indexes = vt.create_neighbours(graph)
order = 2
if regularizing:
regularization = np.ones(hrf_len)
regularization[hrf_len // 3:hrf_len // 2] = 2
regularization[hrf_len // 2:2 * hrf_len // 3] = 5
regularization[2 * hrf_len // 3:3 * hrf_len // 4] = 7
regularization[3 * hrf_len // 4:] = 10
# regularization[hrf_len//2:] = 10
else:
regularization = None
d2 = vt.buildFiniteDiffMatrix(order, hrf_len, regularization)
hrf_regu_prior_inv = d2.T.dot(d2) / pow(dt, 2 * order)
if estimate_hrf and zero_constraint:
hrf_len = hrf_len - 2
hrf_regu_prior_inv = hrf_regu_prior_inv[1:-1, 1:-1]
occurence_matrix = occurence_matrix[:, :, 1:-1]
noise_struct = np.identity(nb_scans)
noise_var = np.ones(nb_voxels)
if nb_classes != 2:
logger.warn('The number of classes is different to two.')
labels_proba = np.zeros((nb_conditions, nb_classes, nb_voxels),
dtype=np.float64)
logger.info("Labels are initialized by setting everything to {}".format(
1. / nb_classes))
labels_proba[:, :, :] = 1. / nb_classes
m_h = getCanoHRF(hrf_duration, dt)[1][:hrf_len]
hrf_mean = np.array(m_h).astype(np.float64)
if estimate_hrf:
hrf_covar = np.identity(hrf_len, dtype=np.float64)
else:
hrf_covar = np.zeros((hrf_len, hrf_len), dtype=np.float64)
beta = beta * np.ones(nb_conditions, dtype=np.float64)
beta_list = [beta.copy()]
if drifts_type == "poly":
drift_basis = vt.poly_drifts_basis(nb_scans, 4, tr)
elif drifts_type == "cos":
drift_basis = vt.cosine_drifts_basis(nb_scans, 64, tr)
else:
raise Exception('drift type "%s" is not supported' % drifts_type)
drift_coeffs = vt.drifts_coeffs_fit(bold_data, drift_basis)
drift = drift_basis.dot(drift_coeffs)
bold_data_drift = bold_data - drift
# Parameters Gaussian mixtures
nrls_class_mean = 2 * np.ones((nb_conditions, nb_classes))
nrls_class_mean[:, 0] = 0
nrls_class_var = 0.3 * np.ones(
(nb_conditions, nb_classes), dtype=np.float64)
nrls_mean = (
np.random.normal(nrls_class_mean, nrls_class_var)[:, :, np.newaxis] *
labels_proba).sum(axis=1).T
nrls_covar = np.identity(nb_conditions)[:, :, np.newaxis] + np.zeros(
(1, 1, nb_voxels))
thresh_free_energy = 1e-4
free_energy = [1.]
free_energy_crit = [1.]
compute_time = []
start_time = time.time()
loop = 0
while (loop <= it_min
or ((np.asarray(free_energy_crit[-5:]) > thresh_free_energy).any()
and loop < it_max)):
logger.info("{:-^80}".format(" Iteration n°" + str(loop + 1) + " "))
logger.info("Expectation A step...")
logger.debug("Before: nrls_mean = %s, nrls_covar = %s", nrls_mean,
nrls_covar)
nrls_mean, nrls_covar = vt.nrls_expectation(
hrf_mean, nrls_mean, occurence_matrix, noise_struct, labels_proba,
nrls_class_mean, nrls_class_var, nb_conditions, bold_data_drift,
nrls_covar, hrf_covar, noise_var)
logger.debug("After: nrls_mean = %s, nrls_covar = %s", nrls_mean,
nrls_covar)
logger.info("Expectation Z step...")
logger.debug("Before: labels_proba = %s, labels_proba = %s",
labels_proba, labels_proba)
labels_proba = vt.labels_expectation(nrls_covar,
nrls_mean,
nrls_class_var,
nrls_class_mean,
beta,
labels_proba,
neighbours_indexes,
nb_conditions,
nb_classes,
nb_voxels,
parallel=True)
logger.debug("After: labels_proba = %s, labels_proba = %s",
labels_proba, labels_proba)
if estimate_hrf:
logger.info("Expectation H step...")
logger.debug("Before: hrf_mean = %s, hrf_covar = %s", hrf_mean,
hrf_covar)
hrf_mean, hrf_covar = vt.hrf_expectation(
nrls_covar, nrls_mean, occurence_matrix, noise_struct,
hrf_regu_prior_inv, sigma_h, nb_voxels, bold_data_drift,
noise_var)
if constrained:
hrf_mean = vt.norm1_constraint(hrf_mean, hrf_covar)
hrf_covar[:] = 0
logger.debug("After: hrf_mean = %s, hrf_covar = %s", hrf_mean,
hrf_covar)
# Normalizing H at each nb_2_norm iterations:
if not constrained and normalizing:
# Normalizing is done before sigma_h, nrls_class_mean and nrls_class_var estimation
# we should not include them in the normalisation step
if (loop + 1) % nb_2_norm == 0:
hrf_norm = np.linalg.norm(hrf_mean)
hrf_mean /= hrf_norm
hrf_covar /= hrf_norm**2
nrls_mean *= hrf_norm
nrls_covar *= hrf_norm**2
if estimate_hrf and estimate_sigma_h:
logger.info("Maximization sigma_H step...")
logger.debug("Before: sigma_h = %s", sigma_h)
if hrf_hyperprior > 0:
sigma_h = vt.maximization_sigmaH_prior(hrf_len, hrf_covar,
hrf_regu_prior_inv,
hrf_mean,
hrf_hyperprior)
else:
sigma_h = vt.maximization_sigmaH(hrf_len, hrf_covar,
hrf_regu_prior_inv, hrf_mean)
logger.debug("After: sigma_h = %s", sigma_h)
logger.info("Maximization (mu,sigma) step...")
logger.debug("Before: nrls_class_mean = %s, nrls_class_var = %s",
nrls_class_mean, nrls_class_var)
nrls_class_mean, nrls_class_var = vt.maximization_class_proba(
labels_proba, nrls_mean, nrls_covar)
logger.debug("After: nrls_class_mean = %s, nrls_class_var = %s",
nrls_class_mean, nrls_class_var)
logger.info("Maximization L step...")
logger.debug("Before: drift_coeffs = %s", drift_coeffs)
drift_coeffs = vt.maximization_drift_coeffs(bold_data, nrls_mean,
occurence_matrix, hrf_mean,
noise_struct, drift_basis)
logger.debug("After: drift_coeffs = %s", drift_coeffs)
drift = drift_basis.dot(drift_coeffs)
bold_data_drift = bold_data - drift
if estimate_beta:
logger.info("Maximization beta step...")
for cond_nb in xrange(0, nb_conditions):
beta[cond_nb], success = vt.beta_maximization(
beta[cond_nb] * np.ones((1, )),
labels_proba[cond_nb, :, :], neighbours_indexes, gamma)
beta_list.append(beta.copy())
logger.debug("beta = %s", str(beta))
logger.info("Maximization sigma noise step...")
noise_var = vt.maximization_noise_var(occurence_matrix, hrf_mean,
hrf_covar, nrls_mean, nrls_covar,
noise_struct, bold_data_drift,
nb_scans)
# Computing Free Energy
free_energy.append(
vt.free_energy_computation(
nrls_mean, nrls_covar, hrf_mean, hrf_covar, hrf_len,
labels_proba, bold_data_drift, occurence_matrix, noise_var,
noise_struct, nb_conditions, nb_voxels, nb_scans, nb_classes,
nrls_class_mean, nrls_class_var,
neighbours_indexes, beta, sigma_h,
np.linalg.inv(hrf_regu_prior_inv), hrf_regu_prior_inv, gamma,
hrf_hyperprior))
free_energy_crit.append(
abs((free_energy[-2] - free_energy[-1]) / free_energy[-2]))
logger.info("Convergence criteria: %f (Threshold = %f)",
free_energy_crit[-1], thresh_free_energy)
loop += 1
compute_time.append(time.time() - start_time)
compute_time_mean = compute_time[-1] / loop
mahalanobis_zero = np.nan
mahalanobis_cano = np.nan
mahalanobis_diff = np.nan
mahalanobis_prod = np.nan
variation_coeff = np.nan
if estimate_hrf and not constrained and not normalizing:
hrf_norm = np.linalg.norm(hrf_mean)
hrf_mean /= hrf_norm
hrf_covar /= hrf_norm**2
sigma_h /= hrf_norm**2
nrls_mean *= hrf_norm
nrls_covar *= hrf_norm**2
nrls_class_mean *= hrf_norm
nrls_class_var *= hrf_norm**2
mahalanobis_zero = mahalanobis(hrf_mean, np.zeros_like(hrf_mean),
np.linalg.inv(hrf_covar))
mahalanobis_cano = mahalanobis(hrf_mean, m_h, np.linalg.inv(hrf_covar))
mahalanobis_diff = mahalanobis_cano - mahalanobis_zero
mahalanobis_prod = mahalanobis_cano * mahalanobis_zero
variation_coeff = np.sqrt((hrf_mean.T.dot(hrf_covar).dot(hrf_mean)) /
(hrf_mean.T.dot(hrf_mean))**2)
if estimate_hrf and zero_constraint:
hrf_mean = np.concatenate(([0], hrf_mean, [0]))
# when using the zero constraint the hrf covariance is fill with
# arbitrary zeros around the matrix, this is maybe a bad idea if we need
# it for later computation...
hrf_covar = np.concatenate((np.zeros(
(hrf_covar.shape[0], 1)), hrf_covar,
np.zeros((hrf_covar.shape[0], 1))),
axis=1)
hrf_covar = np.concatenate((np.zeros(
(1, hrf_covar.shape[1])), hrf_covar,
np.zeros((1, hrf_covar.shape[1]))),
axis=0)
if estimate_hrf:
(delay_of_response, delay_of_undershoot, dispersion_of_response,
dispersion_of_undershoot, ratio_resp_under,
delay) = vt.fit_hrf_two_gammas(hrf_mean, dt, hrf_duration)
else:
(delay_of_response, delay_of_undershoot, dispersion_of_response,
dispersion_of_undershoot, ratio_resp_under,
delay) = (None, None, None, None, None, None)
ppm_a_nrl, ppm_g_nrl = vt.ppms_computation(nrls_mean,
np.diagonal(nrls_covar),
nrls_class_mean,
nrls_class_var,
threshold_a="intersect")
# Calculate contrast maps and variance
nb_contrasts = len(contrasts)
if compute_contrasts and nb_contrasts > 0:
logger.info('Computing contrasts ...')
(contrasts_mean, contrasts_var, contrasts_class_mean,
contrasts_class_var) = vt.contrasts_mean_var_classes(
contrasts, condition_names, nrls_mean, nrls_covar,
nrls_class_mean, nrls_class_var, nb_contrasts, nb_classes,
nb_voxels)
ppm_a_contrasts, ppm_g_contrasts = vt.ppms_computation(
contrasts_mean, contrasts_var, contrasts_class_mean,
contrasts_class_var)
logger.info('Done computing contrasts.')
else:
(contrasts_mean, contrasts_var, contrasts_class_mean,
contrasts_class_var, ppm_a_contrasts,
ppm_g_contrasts) = (None, None, None, None, None, None)
logger.info("Number of iterations to reach criterion: %d", loop)
logger.info("Computational time = {t[0]:.0f} min {t[1]:.0f} s".format(
t=divmod(compute_time[-1], 60)))
logger.debug('nrls_class_mean: %s', nrls_class_mean)
logger.debug('nrls_class_var: %s', nrls_class_var)
logger.debug("sigma_H = %s", str(sigma_h))
logger.debug("beta = %s", str(beta))
stimulus_induced_signal = vt.computeFit(hrf_mean, nrls_mean, X, nb_voxels,
nb_scans)
snr = 20 * np.log(
np.linalg.norm(bold_data.astype(np.float)) / np.linalg.norm(
(bold_data_drift - stimulus_induced_signal).astype(np.float)))
snr /= np.log(10.)
logger.info('SNR comp = %f', snr)
return (loop, nrls_mean, hrf_mean, hrf_covar, labels_proba, noise_var,
nrls_class_mean, nrls_class_var, beta, drift_coeffs, drift,
contrasts_mean, contrasts_var, compute_time[2:], compute_time_mean,
nrls_covar, stimulus_induced_signal, mahalanobis_zero,
mahalanobis_cano, mahalanobis_diff, mahalanobis_prod, ppm_a_nrl,
ppm_g_nrl, ppm_a_contrasts, ppm_g_contrasts, variation_coeff,
free_energy[1:], free_energy_crit[1:], beta_list[1:],
delay_of_response, delay_of_undershoot, dispersion_of_response,
dispersion_of_undershoot, ratio_resp_under, delay)
def Main_vbjde_Extension_constrained_stable(graph, Y, Onsets, Thrf, K, TR, beta,
dt, scale=1, estimateSigmaH=True,
sigmaH=0.05, NitMax=-1,
NitMin=1, estimateBeta=True,
PLOT=False, contrasts=[],
computeContrast=False,
gamma_h=0):
""" Version modified by Lofti from Christine's version """
logger.info(
"Fast EM with C extension started ... Here is the stable version !")
np.random.seed(6537546)
# Initialize parameters
S = 100
if NitMax < 0:
NitMax = 100
gamma = 7.5 # 7.5
gradientStep = 0.003
MaxItGrad = 200
Thresh = 1e-5
# Initialize sizes vectors
D = np.int(np.ceil(Thrf / dt)) + 1
M = len(Onsets)
N = Y.shape[0]
J = Y.shape[1]
l = np.int(np.sqrt(J))
condition_names = []
# Neighbours
maxNeighbours = max([len(nl) for nl in graph])
neighboursIndexes = np.zeros((J, maxNeighbours), dtype=np.int32)
neighboursIndexes -= 1
for i in xrange(J):
neighboursIndexes[i, :len(graph[i])] = graph[i]
# Conditions
X = OrderedDict([])
for condition, Ons in Onsets.iteritems():
X[condition] = vt.compute_mat_X_2(N, TR, D, dt, Ons)
condition_names += [condition]
XX = np.zeros((M, N, D), dtype=np.int32)
nc = 0
for condition, Ons in Onsets.iteritems():
XX[nc, :, :] = X[condition]
nc += 1
# Covariance matrix
order = 2
D2 = vt.buildFiniteDiffMatrix(order, D)
R = np.dot(D2, D2) / pow(dt, 2 * order)
invR = np.linalg.inv(R)
Det_invR = np.linalg.det(invR)
Gamma = np.identity(N)
Det_Gamma = np.linalg.det(Gamma)
Crit_H = 1
Crit_Z = 1
Crit_A = 1
Crit_AH = 1
AH = np.zeros((J, M, D), dtype=np.float64)
AH1 = np.zeros((J, M, D), dtype=np.float64)
Crit_FreeEnergy = 1
cTime = []
cA = []
cH = []
cZ = []
cAH = []
CONTRAST = np.zeros((J, len(contrasts)), dtype=np.float64)
CONTRASTVAR = np.zeros((J, len(contrasts)), dtype=np.float64)
Q_barnCond = np.zeros((M, M, D, D), dtype=np.float64)
XGamma = np.zeros((M, D, N), dtype=np.float64)
m1 = 0
for k1 in X: # Loop over the M conditions
m2 = 0
for k2 in X:
Q_barnCond[m1, m2, :, :] = np.dot(
np.dot(X[k1].transpose(), Gamma), X[k2])
m2 += 1
XGamma[m1, :, :] = np.dot(X[k1].transpose(), Gamma)
m1 += 1
sigma_epsilone = np.ones(J)
logger.info(
"Labels are initialized by setting active probabilities to ones ...")
q_Z = np.zeros((M, K, J), dtype=np.float64)
q_Z[:, 1, :] = 1
q_Z1 = np.zeros((M, K, J), dtype=np.float64)
Z_tilde = q_Z.copy()
TT, m_h = getCanoHRF(Thrf, dt) # TODO: check
m_h = m_h[:D]
m_H = np.array(m_h).astype(np.float64)
m_H1 = np.array(m_h)
sigmaH1 = sigmaH
Sigma_H = np.ones((D, D), dtype=np.float64)
Beta = beta * np.ones((M), dtype=np.float64)
P = vt.PolyMat(N, 4, TR)
L = vt.polyFit(Y, TR, 4, P)
PL = np.dot(P, L)
y_tilde = Y - PL
Ndrift = L.shape[0]
sigma_M = np.ones((M, K), dtype=np.float64)
sigma_M[:, 0] = 0.5
sigma_M[:, 1] = 0.6
mu_M = np.zeros((M, K), dtype=np.float64)
for k in xrange(1, K):
mu_M[:, k] = 1 # InitMean
Sigma_A = np.zeros((M, M, J), np.float64)
for j in xrange(0, J):
Sigma_A[:, :, j] = 0.01 * np.identity(M)
m_A = np.zeros((J, M), dtype=np.float64)
m_A1 = np.zeros((J, M), dtype=np.float64)
for j in xrange(0, J):
for m in xrange(0, M):
for k in xrange(0, K):
m_A[j, m] += np.random.normal(mu_M[m, k],
np.sqrt(sigma_M[m, k])) * q_Z[m, k, j]
m_A1 = m_A
t1 = time.time()
##########################################################################
# VBJDE num. iter. minimum
ni = 0
while ((ni < NitMin + 1) or ((Crit_AH > Thresh) and (ni < NitMax))):
logger.info("------------------------------ Iteration n° " +
str(ni + 1) + " ------------------------------")
#####################
# EXPECTATION
#####################
# A
logger.info("E A step ...")
UtilsC.expectation_A(q_Z, mu_M, sigma_M, PL, sigma_epsilone, Gamma,
Sigma_H, Y, y_tilde, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N, K)
# crit. A
DIFF = np.reshape(m_A - m_A1, (M * J))
Crit_A = (
np.linalg.norm(DIFF) / np.linalg.norm(np.reshape(m_A1, (M * J)))) ** 2
cA += [Crit_A]
m_A1[:, :] = m_A[:, :]
# HRF h
UtilsC.expectation_H(XGamma, Q_barnCond, sigma_epsilone, Gamma, R, Sigma_H, Y,
y_tilde, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N, scale, sigmaH)
#m_H[0] = 0
#m_H[-1] = 0
# Constrain with optimization strategy
import cvxpy as cvx
m, n = Sigma_H.shape
Sigma_H_inv = np.linalg.inv(Sigma_H)
zeros_H = np.zeros_like(m_H[:, np.newaxis])
# Construct the problem. PRIMAL
h = cvx.Variable(n)
expression = cvx.quad_form(h - m_H[:, np.newaxis], Sigma_H_inv)
objective = cvx.Minimize(expression)
#constraints = [h[0] == 0, h[-1]==0, h >= zeros_H, cvx.square(cvx.norm(h,2))<=1]
constraints = [h[0] == 0, h[-1] == 0, cvx.square(cvx.norm(h, 2)) <= 1]
prob = cvx.Problem(objective, constraints)
result = prob.solve(verbose=0, solver=cvx.CVXOPT)
# Now we update the mean of h
m_H_old = m_H
Sigma_H_old = Sigma_H
m_H = np.squeeze(np.array((h.value)))
Sigma_H = np.zeros_like(Sigma_H)
# and the norm
h_norm += [np.linalg.norm(m_H)]
# crit. h
Crit_H = (np.linalg.norm(m_H - m_H1) / np.linalg.norm(m_H1)) ** 2
cH += [Crit_H]
m_H1[:] = m_H[:]
# crit. AH
for d in xrange(0, D):
AH[:, :, d] = m_A[:, :] * m_H[d]
DIFF = np.reshape(AH - AH1, (M * J * D))
Crit_AH = (np.linalg.norm(
DIFF) / (np.linalg.norm(np.reshape(AH1, (M * J * D))) + eps)) ** 2
cAH += [Crit_AH]
AH1[:, :, :] = AH[:, :, :]
# Z labels
logger.info("E Z step ...")
UtilsC.expectation_Z(Sigma_A, m_A, sigma_M, Beta, Z_tilde, mu_M,
q_Z, neighboursIndexes.astype(np.int32), M, J, K, maxNeighbours)
# crit. Z
DIFF = np.reshape(q_Z - q_Z1, (M * K * J))
Crit_Z = (np.linalg.norm(DIFF) /
(np.linalg.norm(np.reshape(q_Z1, (M * K * J))) + eps)) ** 2
cZ += [Crit_Z]
q_Z1[:, :, :] = q_Z[:, :, :]
#####################
# MAXIMIZATION
#####################
# HRF: Sigma_h
if estimateSigmaH:
logger.info("M sigma_H step ...")
if gamma_h > 0:
sigmaH = vt.maximization_sigmaH_prior(
D, Sigma_H, R, m_H, gamma_h)
else:
sigmaH = vt.maximization_sigmaH(D, Sigma_H, R, m_H)
logger.info('sigmaH = %s', str(sigmaH))
# (mu,sigma)
logger.info("M (mu,sigma) step ...")
mu_M, sigma_M = vt.maximization_mu_sigma(
mu_M, sigma_M, q_Z, m_A, K, M, Sigma_A)
# Drift L
UtilsC.maximization_L(
Y, m_A, m_H, L, P, XX.astype(np.int32), J, D, M, Ndrift, N)
PL = np.dot(P, L)
y_tilde = Y - PL
# Beta
if estimateBeta:
logger.info("estimating beta")
for m in xrange(0, M):
Beta[m] = UtilsC.maximization_beta(beta, q_Z[m, :, :].astype(np.float64), Z_tilde[m, :, :].astype(
np.float64), J, K, neighboursIndexes.astype(np.int32), gamma, maxNeighbours, MaxItGrad, gradientStep)
logger.info("End estimating beta")
logger.info(Beta)
# Sigma noise
logger.info("M sigma noise step ...")
UtilsC.maximization_sigma_noise(
Gamma, PL, sigma_epsilone, Sigma_H, Y, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N)
t02 = time.time()
cTime += [t02 - t1]
t2 = time.time()
##########################################################################
# PLOTS and SNR computation
if PLOT and 0:
font = {'size': 15}
matplotlib.rc('font', **font)
savefig('./HRF_Iter_CompMod.png')
hold(False)
figure(2)
plot(cAH[1:-1], 'lightblue')
hold(True)
plot(cFE[1:-1], 'm')
hold(False)
legend(('CAH', 'CFE'))
grid(True)
savefig('./Crit_CompMod.png')
figure(3)
plot(FreeEnergyArray)
grid(True)
savefig('./FreeEnergy_CompMod.png')
figure(4)
for m in xrange(M):
plot(SUM_q_Z_array[m])
hold(True)
hold(False)
savefig('./Sum_q_Z_Iter_CompMod.png')
figure(5)
for m in xrange(M):
plot(mu1_array[m])
hold(True)
hold(False)
savefig('./mu1_Iter_CompMod.png')
figure(6)
plot(h_norm_array)
savefig('./HRF_Norm_CompMod.png')
Data_save = xndarray(h_norm_array, ['Iteration'])
Data_save.save('./HRF_Norm_Comp.nii')
CompTime = t2 - t1
cTimeMean = CompTime / ni
"""
Norm = np.linalg.norm(m_H)
m_H /= Norm
Sigma_H /= Norm**2
sigmaH /= Norm**2
m_A *= Norm
Sigma_A *= Norm**2
mu_M *= Norm
sigma_M *= Norm**2
sigma_M = np.sqrt(np.sqrt(sigma_M))
"""
logger.info("Nb iterations to reach criterion: %d", ni)
logger.info("Computational time = %s min %s s", str(
np.int(CompTime // 60)), str(np.int(CompTime % 60)))
logger.info('mu_M: %f', mu_M)
logger.info('sigma_M: %f', sigma_M)
logger.info("sigma_H = %s", str(sigmaH))
logger.info("Beta = %s", str(Beta))
StimulusInducedSignal = vt.computeFit(m_H, m_A, X, J, N)
SNR = 20 * \
np.log(
np.linalg.norm(Y) / np.linalg.norm(Y - StimulusInducedSignal - PL))
SNR /= np.log(10.)
logger.info('SNR comp = %f', SNR)
return ni, m_A, m_H, q_Z, sigma_epsilone, mu_M, sigma_M, Beta, L, PL, CONTRAST, CONTRASTVAR, cA[2:], cH[2:], cZ[2:], cAH[2:], cTime[2:], cTimeMean, Sigma_A, StimulusInducedSignal
def Main_vbjde_Extension_constrained(graph, Y, Onsets, Thrf, K, TR, beta,
dt, scale=1, estimateSigmaH=True,
sigmaH=0.05, NitMax=-1,
NitMin=1, estimateBeta=True,
PLOT=False, contrasts=[],
computeContrast=False,
gamma_h=0, estimateHRF=True,
TrueHrfFlag=False,
HrfFilename='hrf.nii',
estimateLabels=True,
LabelsFilename='labels.nii',
MFapprox=False, InitVar=0.5,
InitMean=2.0, MiniVEMFlag=False,
NbItMiniVem=5):
# VBJDE Function for BOLD with contraints
logger.info("Fast EM with C extension started ...")
np.random.seed(6537546)
##########################################################################
# INITIALIZATIONS
# Initialize parameters
tau1 = 0.0
tau2 = 0.0
S = 100
Init_sigmaH = sigmaH
Nb2Norm = 1
NormFlag = False
if NitMax < 0:
NitMax = 100
gamma = 7.5
#gamma_h = 1000
gradientStep = 0.003
MaxItGrad = 200
Thresh = 1e-5
Thresh_FreeEnergy = 1e-5
estimateLabels = True # WARNING!! They should be estimated
# Initialize sizes vectors
D = int(np.ceil(Thrf / dt)) + 1 # D = int(np.ceil(Thrf/dt))
M = len(Onsets)
N = Y.shape[0]
J = Y.shape[1]
l = int(np.sqrt(J))
condition_names = []
# Neighbours
maxNeighbours = max([len(nl) for nl in graph])
neighboursIndexes = np.zeros((J, maxNeighbours), dtype=np.int32)
neighboursIndexes -= 1
for i in xrange(J):
neighboursIndexes[i, :len(graph[i])] = graph[i]
# Conditions
X = OrderedDict([])
for condition, Ons in Onsets.iteritems():
X[condition] = vt.compute_mat_X_2(N, TR, D, dt, Ons)
condition_names += [condition]
XX = np.zeros((M, N, D), dtype=np.int32)
nc = 0
for condition, Ons in Onsets.iteritems():
XX[nc, :, :] = X[condition]
nc += 1
# Covariance matrix
order = 2
D2 = vt.buildFiniteDiffMatrix(order, D)
R = np.dot(D2, D2) / pow(dt, 2 * order)
invR = np.linalg.inv(R)
Det_invR = np.linalg.det(invR)
Gamma = np.identity(N)
Det_Gamma = np.linalg.det(Gamma)
p_Wtilde = np.zeros((M, K), dtype=np.float64)
p_Wtilde1 = np.zeros((M, K), dtype=np.float64)
p_Wtilde[:, 1] = 1
Crit_H = 1
Crit_Z = 1
Crit_A = 1
Crit_AH = 1
AH = np.zeros((J, M, D), dtype=np.float64)
AH1 = np.zeros((J, M, D), dtype=np.float64)
Crit_FreeEnergy = 1
cA = []
cH = []
cZ = []
cAH = []
FreeEnergy_Iter = []
cTime = []
cFE = []
SUM_q_Z = [[] for m in xrange(M)]
mu1 = [[] for m in xrange(M)]
h_norm = []
h_norm2 = []
CONTRAST = np.zeros((J, len(contrasts)), dtype=np.float64)
CONTRASTVAR = np.zeros((J, len(contrasts)), dtype=np.float64)
Q_barnCond = np.zeros((M, M, D, D), dtype=np.float64)
XGamma = np.zeros((M, D, N), dtype=np.float64)
m1 = 0
for k1 in X: # Loop over the M conditions
m2 = 0
for k2 in X:
Q_barnCond[m1, m2, :, :] = np.dot(
np.dot(X[k1].transpose(), Gamma), X[k2])
m2 += 1
XGamma[m1, :, :] = np.dot(X[k1].transpose(), Gamma)
m1 += 1
if MiniVEMFlag:
logger.info("MiniVEM to choose the best initialisation...")
"""InitVar, InitMean, gamma_h = MiniVEM_CompMod(Thrf,TR,dt,beta,Y,K,
gamma,gradientStep,
MaxItGrad,D,M,N,J,S,
maxNeighbours,
neighboursIndexes,
XX,X,R,Det_invR,Gamma,
Det_Gamma,
scale,Q_barnCond,XGamma,
NbItMiniVem,
sigmaH,estimateHRF)"""
InitVar, InitMean, gamma_h = vt.MiniVEM_CompMod(Thrf, TR, dt, beta, Y, K, gamma, gradientStep, MaxItGrad, D, M, N, J, S, maxNeighbours,
neighboursIndexes, XX, X, R, Det_invR, Gamma, Det_Gamma, p_Wtilde, scale, Q_barnCond, XGamma, tau1, tau2, NbItMiniVem, sigmaH, estimateHRF)
sigmaH = Init_sigmaH
sigma_epsilone = np.ones(J)
logger.info(
"Labels are initialized by setting active probabilities to ones ...")
q_Z = np.zeros((M, K, J), dtype=np.float64)
q_Z[:, 1, :] = 1
q_Z1 = np.zeros((M, K, J), dtype=np.float64)
Z_tilde = q_Z.copy()
# TT,m_h = getCanoHRF(Thrf-dt,dt) #TODO: check
TT, m_h = getCanoHRF(Thrf, dt) # TODO: check
m_h = m_h[:D]
m_H = np.array(m_h).astype(np.float64)
m_H1 = np.array(m_h)
sigmaH1 = sigmaH
if estimateHRF:
Sigma_H = np.ones((D, D), dtype=np.float64)
else:
Sigma_H = np.zeros((D, D), dtype=np.float64)
Beta = beta * np.ones((M), dtype=np.float64)
P = vt.PolyMat(N, 4, TR)
L = vt.polyFit(Y, TR, 4, P)
PL = np.dot(P, L)
y_tilde = Y - PL
Ndrift = L.shape[0]
sigma_M = np.ones((M, K), dtype=np.float64)
sigma_M[:, 0] = 0.5
sigma_M[:, 1] = 0.6
mu_M = np.zeros((M, K), dtype=np.float64)
for k in xrange(1, K):
mu_M[:, k] = InitMean
Sigma_A = np.zeros((M, M, J), np.float64)
for j in xrange(0, J):
Sigma_A[:, :, j] = 0.01 * np.identity(M)
m_A = np.zeros((J, M), dtype=np.float64)
m_A1 = np.zeros((J, M), dtype=np.float64)
for j in xrange(0, J):
for m in xrange(0, M):
for k in xrange(0, K):
m_A[j, m] += np.random.normal(mu_M[m, k],
np.sqrt(sigma_M[m, k])) * q_Z[m, k, j]
m_A1 = m_A
t1 = time.time()
##########################################################################
# VBJDE num. iter. minimum
ni = 0
while ((ni < NitMin) or (((Crit_FreeEnergy > Thresh_FreeEnergy) or (Crit_AH > Thresh)) and (ni < NitMax))):
logger.info("------------------------------ Iteration n° " +
str(ni + 1) + " ------------------------------")
#####################
# EXPECTATION
#####################
# A
logger.info("E A step ...")
UtilsC.expectation_A(q_Z, mu_M, sigma_M, PL, sigma_epsilone, Gamma,
Sigma_H, Y, y_tilde, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N, K)
val = np.reshape(m_A, (M * J))
val[np.where((val <= 1e-50) & (val > 0.0))] = 0.0
val[np.where((val >= -1e-50) & (val < 0.0))] = 0.0
# crit. A
DIFF = np.reshape(m_A - m_A1, (M * J))
# To avoid numerical problems
DIFF[np.where((DIFF < 1e-50) & (DIFF > 0.0))] = 0.0
# To avoid numerical problems
DIFF[np.where((DIFF > -1e-50) & (DIFF < 0.0))] = 0.0
Crit_A = (
np.linalg.norm(DIFF) / np.linalg.norm(np.reshape(m_A1, (M * J)))) ** 2
cA += [Crit_A]
m_A1[:, :] = m_A[:, :]
# HRF h
if estimateHRF:
################################
# HRF ESTIMATION
################################
UtilsC.expectation_H(XGamma, Q_barnCond, sigma_epsilone, Gamma, R, Sigma_H, Y,
y_tilde, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N, scale, sigmaH)
import cvxpy as cvx
m, n = Sigma_H.shape
Sigma_H_inv = np.linalg.inv(Sigma_H)
zeros_H = np.zeros_like(m_H[:, np.newaxis])
# Construct the problem. PRIMAL
h = cvx.Variable(n)
expression = cvx.quad_form(h - m_H[:, np.newaxis], Sigma_H_inv)
objective = cvx.Minimize(expression)
#constraints = [h[0] == 0, h[-1]==0, h >= zeros_H, cvx.square(cvx.norm(h,2))<=1]
constraints = [
h[0] == 0, h[-1] == 0, cvx.square(cvx.norm(h, 2)) <= 1]
prob = cvx.Problem(objective, constraints)
result = prob.solve(verbose=0, solver=cvx.CVXOPT)
# Now we update the mean of h
m_H_old = m_H
Sigma_H_old = Sigma_H
m_H = np.squeeze(np.array((h.value)))
Sigma_H = np.zeros_like(Sigma_H)
h_norm += [np.linalg.norm(m_H)]
# print 'h_norm = ', h_norm
# Plotting HRF
if PLOT and ni >= 0:
import matplotlib.pyplot as plt
plt.figure(M + 1)
plt.plot(m_H)
plt.hold(True)
else:
if TrueHrfFlag:
#TrueVal, head = read_volume(HrfFilename)
TrueVal, head = read_volume(HrfFilename)[:, 0, 0, 0]
print TrueVal
print TrueVal.shape
m_H = TrueVal
# crit. h
Crit_H = (np.linalg.norm(m_H - m_H1) / np.linalg.norm(m_H1)) ** 2
cH += [Crit_H]
m_H1[:] = m_H[:]
# crit. AH
for d in xrange(0, D):
AH[:, :, d] = m_A[:, :] * m_H[d]
DIFF = np.reshape(AH - AH1, (M * J * D))
# To avoid numerical problems
DIFF[np.where((DIFF < 1e-50) & (DIFF > 0.0))] = 0.0
# To avoid numerical problems
DIFF[np.where((DIFF > -1e-50) & (DIFF < 0.0))] = 0.0
if np.linalg.norm(np.reshape(AH1, (M * J * D))) == 0:
Crit_AH = 1000000000.
else:
Crit_AH = (
np.linalg.norm(DIFF) / np.linalg.norm(np.reshape(AH1, (M * J * D)))) ** 2
cAH += [Crit_AH]
AH1[:, :, :] = AH[:, :, :]
# Z labels
if estimateLabels:
logger.info("E Z step ...")
# WARNING!!! ParsiMod gives better results, but we need the other
# one.
if MFapprox:
UtilsC.expectation_Z(Sigma_A, m_A, sigma_M, Beta, Z_tilde, mu_M, q_Z, neighboursIndexes.astype(
np.int32), M, J, K, maxNeighbours)
if not MFapprox:
UtilsC.expectation_Z_ParsiMod_RVM_and_CompMod(
Sigma_A, m_A, sigma_M, Beta, mu_M, q_Z, neighboursIndexes.astype(np.int32), M, J, K, maxNeighbours)
else:
logger.info("Using True Z ...")
TrueZ = read_volume(LabelsFilename)
for m in xrange(M):
q_Z[m, 1, :] = np.reshape(TrueZ[0][:, :, :, m], J)
q_Z[m, 0, :] = 1 - q_Z[m, 1, :]
# crit. Z
val = np.reshape(q_Z, (M * K * J))
val[np.where((val <= 1e-50) & (val > 0.0))] = 0.0
DIFF = np.reshape(q_Z - q_Z1, (M * K * J))
# To avoid numerical problems
DIFF[np.where((DIFF < 1e-50) & (DIFF > 0.0))] = 0.0
# To avoid numerical problems
DIFF[np.where((DIFF > -1e-50) & (DIFF < 0.0))] = 0.0
if np.linalg.norm(np.reshape(q_Z1, (M * K * J))) == 0:
Crit_Z = 1000000000.
else:
Crit_Z = (
np.linalg.norm(DIFF) / np.linalg.norm(np.reshape(q_Z1, (M * K * J)))) ** 2
cZ += [Crit_Z]
q_Z1 = q_Z
#####################
# MAXIMIZATION
#####################
# HRF: Sigma_h
if estimateHRF:
if estimateSigmaH:
logger.info("M sigma_H step ...")
if gamma_h > 0:
sigmaH = vt.maximization_sigmaH_prior(
D, Sigma_H_old, R, m_H_old, gamma_h)
else:
sigmaH = vt.maximization_sigmaH(D, Sigma_H, R, m_H)
logger.info('sigmaH = %s', str(sigmaH))
# (mu,sigma)
logger.info("M (mu,sigma) step ...")
mu_M, sigma_M = vt.maximization_mu_sigma(
mu_M, sigma_M, q_Z, m_A, K, M, Sigma_A)
for m in xrange(M):
SUM_q_Z[m] += [sum(q_Z[m, 1, :])]
mu1[m] += [mu_M[m, 1]]
# Drift L
UtilsC.maximization_L(
Y, m_A, m_H, L, P, XX.astype(np.int32), J, D, M, Ndrift, N)
PL = np.dot(P, L)
y_tilde = Y - PL
# Beta
if estimateBeta:
logger.info("estimating beta")
for m in xrange(0, M):
if MFapprox:
Beta[m] = UtilsC.maximization_beta(beta, q_Z[m, :, :].astype(np.float64), Z_tilde[m, :, :].astype(
np.float64), J, K, neighboursIndexes.astype(np.int32), gamma, maxNeighbours, MaxItGrad, gradientStep)
if not MFapprox:
#Beta[m] = UtilsC.maximization_beta(beta,q_Z[m,:,:].astype(np.float64),q_Z[m,:,:].astype(np.float64),J,K,neighboursIndexes.astype(int32),gamma,maxNeighbours,MaxItGrad,gradientStep)
Beta[m] = UtilsC.maximization_beta_CB(beta, q_Z[m, :, :].astype(
np.float64), J, K, neighboursIndexes.astype(np.int32), gamma, maxNeighbours, MaxItGrad, gradientStep)
logger.info("End estimating beta")
logger.info(Beta)
# Sigma noise
logger.info("M sigma noise step ...")
UtilsC.maximization_sigma_noise(
Gamma, PL, sigma_epsilone, Sigma_H, Y, m_A, m_H, Sigma_A, XX.astype(np.int32), J, D, M, N)
#### Computing Free Energy ####
if ni > 0:
FreeEnergy1 = FreeEnergy
"""FreeEnergy = vt.Compute_FreeEnergy(y_tilde,m_A,Sigma_A,mu_M,sigma_M,
m_H,Sigma_H,R,Det_invR,sigmaH,
p_Wtilde,q_Z,neighboursIndexes,
maxNeighbours,Beta,sigma_epsilone,
XX,Gamma,Det_Gamma,XGamma,J,D,M,
N,K,S,"CompMod")"""
FreeEnergy = vt.Compute_FreeEnergy(y_tilde, m_A, Sigma_A, mu_M, sigma_M, m_H, Sigma_H, R, Det_invR, sigmaH, p_Wtilde, tau1,
tau2, q_Z, neighboursIndexes, maxNeighbours, Beta, sigma_epsilone, XX, Gamma, Det_Gamma, XGamma, J, D, M, N, K, S, "CompMod")
if ni > 0:
Crit_FreeEnergy = (FreeEnergy1 - FreeEnergy) / FreeEnergy1
FreeEnergy_Iter += [FreeEnergy]
cFE += [Crit_FreeEnergy]
# Update index
ni += 1
t02 = time.time()
cTime += [t02 - t1]
t2 = time.time()
##########################################################################
# PLOTS and SNR computation
FreeEnergyArray = np.zeros((ni), dtype=np.float64)
for i in xrange(ni):
FreeEnergyArray[i] = FreeEnergy_Iter[i]
SUM_q_Z_array = np.zeros((M, ni), dtype=np.float64)
mu1_array = np.zeros((M, ni), dtype=np.float64)
h_norm_array = np.zeros((ni), dtype=np.float64)
for m in xrange(M):
for i in xrange(ni):
SUM_q_Z_array[m, i] = SUM_q_Z[m][i]
mu1_array[m, i] = mu1[m][i]
h_norm_array[i] = h_norm[i]
if PLOT and 0:
import matplotlib.pyplot as plt
import matplotlib
font = {'size': 15}
matplotlib.rc('font', **font)
plt.savefig('./HRF_Iter_CompMod.png')
plt.hold(False)
plt.figure(2)
plt.plot(cAH[1:-1], 'lightblue')
plt.hold(True)
plt.plot(cFE[1:-1], 'm')
plt.hold(False)
#plt.legend( ('CA','CH', 'CZ', 'CAH', 'CFE') )
plt.legend(('CAH', 'CFE'))
plt.grid(True)
plt.savefig('./Crit_CompMod.png')
plt.figure(3)
plt.plot(FreeEnergyArray)
plt.grid(True)
plt.savefig('./FreeEnergy_CompMod.png')
plt.figure(4)
for m in xrange(M):
plt.plot(SUM_q_Z_array[m])
plt.hold(True)
plt.hold(False)
#plt.legend( ('m=0','m=1', 'm=2', 'm=3') )
#plt.legend( ('m=0','m=1') )
plt.savefig('./Sum_q_Z_Iter_CompMod.png')
plt.figure(5)
for m in xrange(M):
plt.plot(mu1_array[m])
plt.hold(True)
plt.hold(False)
plt.savefig('./mu1_Iter_CompMod.png')
plt.figure(6)
plt.plot(h_norm_array)
plt.savefig('./HRF_Norm_CompMod.png')
Data_save = xndarray(h_norm_array, ['Iteration'])
Data_save.save('./HRF_Norm_Comp.nii')
CompTime = t2 - t1
cTimeMean = CompTime / ni
sigma_M = np.sqrt(np.sqrt(sigma_M))
logger.info("Nb iterations to reach criterion: %d", ni)
logger.info("Computational time = %s min %s s", str(
int(CompTime // 60)), str(int(CompTime % 60)))
# print "Computational time = " + str(int( CompTime//60 ) ) + " min " + str(int(CompTime%60)) + " s"
# print "sigma_H = " + str(sigmaH)
logger.info('mu_M: %f', mu_M)
logger.info('sigma_M: %f', sigma_M)
logger.info("sigma_H = %s" + str(sigmaH))
logger.info("Beta = %s" + str(Beta))
StimulusInducedSignal = vt.computeFit(m_H, m_A, X, J, N)
SNR = 20 * \
np.log(
np.linalg.norm(Y) / np.linalg.norm(Y - StimulusInducedSignal - PL))
SNR /= np.log(10.)
logger.info("SNR = %d", SNR)
return ni, m_A, m_H, q_Z, sigma_epsilone, mu_M, sigma_M, Beta, L, PL, CONTRAST, CONTRASTVAR, cA[2:], cH[2:], cZ[2:], cAH[2:], cTime[2:], cTimeMean, Sigma_A, StimulusInducedSignal, FreeEnergyArray
