eqn.objective = BayesianObjectiveStats(qtarget, beta_prior, sigma_obs, sigma_prior)
n = np.size(qtarget)/6
try:
dat = io.loadmat("H.mat")
H = data["H"]
print 1000*"#"
print "Using Hessian from file"
except:
njac = 10
Jac = np.zeros([njac, 4*n])
for i in range(njac):
print "\rBuilding jac %i of %i"%(i, 5*n),
sys.stdout.flush()
eqn.objective.set_index(i)
dJdbeta = eqn.calc_sensitivity()
Jac[i,:] = dJdbeta[:]
H = Jac.transpose().dot(Jac)/sigma_obs**2 + np.eye(5*n)/sigma_prior**2
io.savemat("H.mat", {"H":H})
Cov = linalg.inv(H)
R = linalg.cholesky(Cov)
Q_post = np.zeros([np.size(eqn.q), nsamples])
sample = 0
while sample < nsamples:
print "\rSampling post %i of %i"%(sample, nsamples),
sys.stdout.flush()
eqn.dt = dt
eqn.q[:] = eqn_map.q[:]
eqn.beta[:] = beta_map + R.dot(np.random.randn(np.shape(beta_map)))
