def generate_smooth_gp_re_a(out_fname='data.csv', country_variation=True):
""" Generate random data based on a nested gaussian process random
effects model with age, with covariates that vary smoothly over
time (where unexplained variation in time does not interact with
unexplained variation in age)
This function generates data for all countries in all regions, and
all age groups based on the model::
Y_r,c,t = beta * X_r,c,t + f_r(t) + g_r(a) + f_c(t)
beta = [30., -.5, .1, .1, -.1, 0., 0., 0., 0., 0.]
f_r ~ GP(0, C(3.))
g_r ~ GP(0, C(2.))
f_c ~ GP(0, C(1.)) or 0 depending on country_variation flag
C(amp) = Matern(amp, scale=20., diff_degree=2)
X_r,c,t[0] = 1
X_r,c,t[1] = t - 1990.
X_r,c,t[k] ~ GP(t; 0, C(1)) for k >= 2
"""
c4 = countries_by_region()
data = col_names()
beta = [30., -.5, .1, .1, -.1, 0., 0., 0., 0., 0.]
C0 = gp.matern.euclidean(time_range, time_range, amp=1., scale=25., diff_degree=2)
C1 = gp.matern.euclidean(age_range, age_range, amp=1., scale=25., diff_degree=2)
C2 = gp.matern.euclidean(time_range, time_range, amp=.1, scale=25., diff_degree=2)
C3 = gp.matern.euclidean(time_range, time_range, amp=1., scale=25., diff_degree=2)
g = mc.rmv_normal_cov(pl.zeros_like(age_range), C1)
for r in c4:
f_r = mc.rmv_normal_cov(pl.zeros_like(time_range), C0)
g_r = mc.rmv_normal_cov(g, C1)
for c in c4[r]:
f_c = mc.rmv_normal_cov(pl.zeros_like(time_range), C2)
x_gp = {}
for k in range(2,10):
x_gp[k] = mc.rmv_normal_cov(pl.zeros_like(time_range), C3)
for j, t in enumerate(time_range):
for i, a in enumerate(age_range):
x = [1] + [j] + [x_gp[k][j] for k in range(2,10)]
y = float(pl.dot(beta, x)) + f_r[j] + g_r[i]
if country_variation:
y += f_c[j]
se = 0.
data.append([r, c, t, a, y, se] + list(x))
write(data, out_fname)
def generate_smooth_gp_re_a(out_fname='data.csv', country_variation=True):
""" Generate random data based on a nested gaussian process random
effects model with age, with covariates that vary smoothly over
time (where unexplained variation in time does not interact with
unexplained variation in age)
This function generates data for all countries in all regions, and
all age groups based on the model::
Y_r,c,t = beta * X_r,c,t + f_r(t) + g_r(a) + f_c(t)
beta = [30., -.5, .1, .1, -.1, 0., 0., 0., 0., 0.]
f_r ~ GP(0, C(3.))
g_r ~ GP(0, C(2.))
f_c ~ GP(0, C(1.)) or 0 depending on country_variation flag
C(amp) = Matern(amp, scale=20., diff_degree=2)
X_r,c,t[0] = 1
X_r,c,t[1] = t - 1990.
X_r,c,t[k] ~ GP(t; 0, C(1)) for k >= 2
"""
c4 = countries_by_region()
data = col_names()
beta = [30., -.5, .1, .1, -.1, 0., 0., 0., 0., 0.]
C0 = gp.matern.euclidean(time_range, time_range, amp=1., scale=25., diff_degree=2)
C1 = gp.matern.euclidean(age_range, age_range, amp=1., scale=25., diff_degree=2)
C2 = gp.matern.euclidean(time_range, time_range, amp=.1, scale=25., diff_degree=2)
C3 = gp.matern.euclidean(time_range, time_range, amp=1., scale=25., diff_degree=2)
g = mc.rmv_normal_cov(pl.zeros_like(age_range), C1)
for r in c4:
f_r = mc.rmv_normal_cov(pl.zeros_like(time_range), C0)
g_r = mc.rmv_normal_cov(g, C1)
for c in c4[r]:
f_c = mc.rmv_normal_cov(pl.zeros_like(time_range), C2)
x_gp = {}
for k in range(2,10):
x_gp[k] = mc.rmv_normal_cov(pl.zeros_like(time_range), C3)
for j, t in enumerate(time_range):
for i, a in enumerate(age_range):
x = [1] + [j] + [x_gp[k][j] for k in range(2,10)]
y = float(pl.dot(beta, x)) + f_r[j] + g_r[i]
if country_variation:
y += f_c[j]
se = 0.
data.append([r, c, t, a, y, se] + list(x))
write(data, out_fname)
def propose(self):
ee = np.asarray(self.Y.value) - np.asarray(self.muY.value)
H = (np.exp(self.LH.value)**(-0.5))[1:]
K = np.asarray(self.Y.value).shape[1]
b_new = np.empty_like(self.stochastic.value)
# auxiliary variables to pick the right subvector/submatrix for the equations
lb = 0
ub = 1
for j in range(1, K):
z = np.expand_dims(H[:, j], 1)*np.expand_dims(ee[:, j], 1) # LHS variable in the regression
Z = np.expand_dims(-H[:, j], 1)*ee[:, :j] # RHS variables in the regression
b_prior = np.asarray([self.b_bar[lb:ub]])
Vinv_prior = inv(self.Pb_bar[lb:ub, lb:ub])
V_post = inv(Vinv_prior + Z.T @ Z)
b_post = V_post @ (Vinv_prior @ b_prior.T + Z.T @ z)
b_new[lb:ub] = pm.rmv_normal_cov(b_post.ravel(), V_post)
lb = ub
ub += j+1
self.stochastic.value = b_new
def _set_initial_values(self, alpha0, nu0, Phi0, mu0, Sigma0, weights0,
alpha_a0, alpha_b0):
if nu0 is None:
nu0 = 3
if Phi0 is None:
Phi0 = np.empty((self.ncomp, self.ndim, self.ndim))
Phi0[:] = np.eye(self.ndim) * (nu0 - 1)
if Sigma0 is None:
# draw from prior
Sigma0 = np.empty((self.ncomp, self.ndim, self.ndim))
for j in xrange(self.ncomp):
Sigma0[j] = pm.rinverse_wishart_prec(nu0 + 2 + self.ncomp, Phi0[j])
# starting values, are these sensible?
if mu0 is None:
mu0 = np.empty((self.ncomp, self.ndim))
for j in xrange(self.ncomp):
mu0[j] = pm.rmv_normal_cov(self.mu_prior_mean,
self.gamma[j] * Sigma0[j])
if weights0 is None:
_, weights0 = stick_break_proc(1, 1, size=self.ncomp - 1)
self._alpha0 = alpha0
self._alpha_a0 = alpha_a0
self._alpha_b0 = alpha_b0
self._weights0 = weights0
self._mu0 = mu0
self._Sigma0 = Sigma0
self._nu0 = nu0 # prior degrees of freedom
self._Phi0 = Phi0 # prior location for Sigma_j's
def complex_hierarchical_data(n):
""" Generate data based on the much more complicated model
given in section 3.2.1::
y_ij ~ N(mu_j - exp(beta_j)t_ij - exp(gamma_j)t_ij^2, sigma_j^2)
gamma_j | sigma^2, xi, X_j ~ N(eta_0 + eta_1 X_j + eta_2 X_j^2, omega^2)
beta_j | gamma_j, sigma^2, xi, X_j ~ N(delta_beta_0 + delta_beta_1 X_j + delta_beta_2 X_j^2 + delta_beta_3 gamma_j, omega_beta^2)
mu_j | gamma_j, beta_j, sigma^2, xi, X_j ~ N(delta_mu_0 + delta_mu_1 X_j + delta_mu_2 X_j^2 + delta_mu_3 gamma_j + delta_mu_4 beta_j, omega_mu^2)
eta = (eta_0, eta_1, eta_2, log(omega))'
delta_beta = (delta_beta_0, delta_beta_1, delta_beta_2, delta_beta_3, log(omega_beta))'
delta_mu = (delta_mu_0, delta_mu_1, delta_mu_2, delta_mu_3, log(omega_mu))'
xi = (eta, delta_beta, delta_mu)
eta ~ MVNormal(M, C)
delta_beta, delta_mu ~ Normal(m, s)
Parameters
----------
n : list, len(n) = J, n[j] = num observations in group j
"""
J = len(n)
# covariate data, not entirely specified in paper
X = mc.rnormal(0, .1**-2, size=J)
t = [pl.arange(n[j]) for j in range(J)]
# hyper-priors, not specified in detail in paper
m = 0.
s = 1.
M = pl.zeros(4)
r = [[ 1, .57, .18, .56],
[.57, 1, .72, .16],
[.18, .72, 1, .14],
[.56, .16, .14, 1]]
eta = mc.rmv_normal_cov(M, r)
omega = .0001 #pl.exp(eta[-1])
delta_beta = mc.rnormal(m, s**-2, size=5)
omega_beta = .0001 #pl.exp(delta_beta[-1])
delta_mu = mc.rnormal(m, s**-2, size=5)
omega_mu = .0001 #pl.exp(delta_mu[-1])
gamma = mc.rnormal(eta[0] + eta[1]*X + eta[2]*X**2, omega**-2.)
beta = mc.rnormal(delta_beta[0] + delta_beta[1]*X + delta_beta[2]*X**2 + delta_beta[3]*gamma, omega_beta**-2)
mu = mc.rnormal(delta_mu[0] + delta_mu[1]*X + delta_mu[2]*X**2 + delta_mu[3]*gamma + delta_mu[4]*beta, omega_mu**-2)
# stochastic error, not specified in paper
sigma = .01*pl.ones(J)
y = [mc.rnormal(mu[j] - pl.exp(beta[j])*t[j] - pl.exp(gamma[j])*t[j]**2, sigma[j]**-2) for j in range(J)]
eta_cross_eta = [eta[0]*eta[1], eta[0]*eta[2], eta[0]*eta[3], eta[1]*eta[2], eta[1]*eta[2], eta[2]*eta[3]]
return vars()
def simulate_data(M_pri, C_pri, N_samp, V, N_exam, N_age_samps, correction_factor_array, age_lims):
"""Called by pred_samps in the outer loop to simulate data."""
# Draw P' from prior.
f_samp = pm.rmv_normal_cov(M_pri, C_pri + eye(N_samp)*V)
# Get ages, number positive, and normalized age distribution for prediction
ages, positives, age_distribution = ages_and_data(N_exam, f_samp, correction_factor_array, age_lims)
sig = sqrt(diag(C_pri))
lo = M_pri - sig*5
hi = M_pri + sig*5
# Make log-likelihood functions
marginal_log_likelihoods = known_age_corr_likelihoods_f(positives, ages, correction_factor_array, linspace(lo.min(),hi.max(),500), 0)
return marginal_log_likelihoods, positives
def step(self):
pri_sig = np.asarray(self.sig.value)
lo = pm.gp.trisolve(pri_sig, self.x.T, uplo='L').T
post_tau = np.dot(lo,lo.T)
l = np.linalg.cholesky(post_tau)
post_C = pm.gp.trisolve(l, np.eye(l.shape[0]),uplo='L')
post_C = pm.gp.trisolve(l.T, post_C, uplo='U')
post_mean = np.dot(lo, pm.gp.trisolve(pri_sig, self.d, uplo='L'))
post_mean = pm.gp.trisolve(l, post_mean, uplo='L')
post_mean = pm.gp.trisolve(l.T, post_mean, uplo='U')
new_val = pm.rmv_normal_cov(post_mean, post_C).squeeze()
[b.set_value(nv) for (b,nv) in zip(self.beta, new_val)]
def simulate_data(M_pri, C_pri, N_samp, V, N_exam, N_age_samps,
correction_factor_array, age_lims):
"""Called by pred_samps in the outer loop to simulate data."""
# Draw P' from prior.
f_samp = pm.rmv_normal_cov(M_pri, C_pri + eye(N_samp) * V)
# Get ages, number positive, and normalized age distribution for prediction
ages, positives, age_distribution = ages_and_data(N_exam, f_samp,
correction_factor_array,
age_lims)
sig = sqrt(diag(C_pri))
lo = M_pri - sig * 5
hi = M_pri + sig * 5
# Make log-likelihood functions
marginal_log_likelihoods = known_age_corr_likelihoods_f(
positives, ages, correction_factor_array,
linspace(lo.min(), hi.max(), 500), 0)
return marginal_log_likelihoods, positives
def _update_mu_Sigma(self, Sigma, component_mask):
mu_output = np.zeros((self.ncomp, self.ndim))
Sigma_output = np.zeros((self.ncomp, self.ndim, self.ndim))
for j in xrange(self.ncomp):
mask = component_mask[j]
Xj = self.data[mask]
nj = len(Xj)
# TODO: sample from prior if nj == 0
sumxj = Xj.sum(0)
gam = self.gamma[j]
mu_hyper = self.mu_prior_mean
post_mean = (mu_hyper / gam + sumxj) / (1 / gam + nj)
post_cov = 1 / (1 / gam + nj) * Sigma[j]
new_mu = pm.rmv_normal_cov(post_mean, post_cov)
Xj_demeaned = Xj - new_mu
mu_SS = np.outer(new_mu - mu_hyper, new_mu - mu_hyper) / gam
data_SS = np.dot(Xj_demeaned.T, Xj_demeaned)
post_Phi = data_SS + mu_SS + self._nu0 * self._Phi0[j]
# symmetrize
post_Phi = (post_Phi + post_Phi.T) / 2
# P(Sigma) ~ IW(nu + 2, nu * Phi)
# P(Sigma | theta, Y) ~
post_nu = nj + self.ncomp + self._nu0 + 3
# pymc rinverse_wishart takes
new_Sigma = pm.rinverse_wishart_prec(post_nu, post_Phi)
mu_output[j] = new_mu
Sigma_output[j] = new_Sigma
return mu_output, Sigma_output
def generate_data(n=1e5, k=2, ncomps=3, seed=1):
npr.seed(seed)
data_concat = []
labels_concat = []
for j in xrange(ncomps):
mean = gen_mean[j]
sd = gen_sd[j]
corr = gen_corr[j]
cov = np.empty((k, k))
cov.fill(corr)
cov[np.diag_indices(k)] = 1
cov *= np.outer(sd, sd)
num = int(n * group_weights[j])
rvs = pm.rmv_normal_cov(mean, cov, size=num)
data_concat.append(rvs)
labels_concat.append(np.repeat(j, num))
return (np.concatenate(labels_concat), np.concatenate(data_concat, axis=0))
def generate_data(n=1e5, k=2, ncomps=3, seed=1):
npr.seed(seed)
data_concat = []
labels_concat = []
for j in xrange(ncomps):
mean = gen_mean[j]
sd = gen_sd[j]
corr = gen_corr[j]
cov = np.empty((k, k))
cov.fill(corr)
cov[np.diag_indices(k)] = 1
cov *= np.outer(sd, sd)
num = int(n * group_weights[j])
rvs = pm.rmv_normal_cov(mean, cov, size=num)
data_concat.append(rvs)
labels_concat.append(np.repeat(j, num))
return (np.concatenate(labels_concat), np.concatenate(data_concat, axis=0))
def sample_covariates(covariate_dict, C_eval, d):
"""
Samples covariates back in when they have been marginalized away.
- covariate_dict : {name : value-on-input, prior-variance}
- M_eval : array. Probably zeros, unless you did something fancy in the mean.
- C_eval : covariance of d | covariates, m
- d : current deviation from mean of covariates' immediate child.
"""
# Extract keys to list to preserve order.
n = covariate_dict.keys()
cvv = [covariate_dict[k] for k in n]
x = np.asarray([v[0] for v in cvv])
prior_var = np.diag([v[1] for v in cvv])
prior_offdiag = np.dot(prior_var,x).T
prior_S = np.linalg.cholesky(np.asarray(C_eval) + np.dot(prior_offdiag, x))
pm.gp.trisolve(prior_S, prior_offdiag, uplo='L', transa='N', inplace=True)
post_C = prior_var - np.dot(prior_offdiag.T, prior_offdiag)
post_mean = np.dot(prior_offdiag.T, pm.gp.trisolve(prior_S, d, uplo='L', transa='N'))
new_val = pm.rmv_normal_cov(post_mean, post_C).squeeze()
return dict(zip(n, new_val))
[ \rho\sigma_2\sigma_1 \sigma_2\sigma_2 ]
Then, knowing the covariance matrix C and given the random samples,
we want to estimate the posterior distribution of mu. Since the prior
for mu is uniform, the mean posterior distribution is simply a bivariate
normal with the same correlation coefficient rho, but with variances
divided by sqrt(N), where N is the number of samples drawn.
We can check that the sampler works correctly by making sure that
after a while, the covariance matrix of the samples for mu tend to C/N.
"""
N = 50
mu = np.array([-2., 3.])
C = np.array([[1, .8 * np.sqrt(2)], [.8 * np.sqrt(2), 2.]])
r = pymc.rmv_normal_cov(mu, C, size=50)
@pymc.stoch
def mean(value=np.array([0., 0.])):
"""The mean of the samples (mu). """
return 0.
obs = pymc.MvNormalCov('obs', mean, C, value=r, observed=True)
class TestAM(TestCase):
def test_convergence(self):
S = pymc.MCMC([mean, obs])
S.use_step_method(pymc.AdaptiveMetropolis, mean, delay=200)
t=np.random.normal(size=n_data+n_pred)
cv = {}
if len(names)>2:
for name in names[:-2]:
cv[name] = np.random.normal(size=n_data+n_pred)*on#np.ones(n_data)
cv['m'] = np.ones(n_data+n_pred)*on
cv['t'] = t*on
C = pm.gp.FullRankCovariance(my_st, amp=1, scale=1, inc=np.pi/4, ecc=.3,st=.1, sd=.5, tlc=.2, sf = .1)
dm = np.vstack((lon,lat,t)).T
C_eval = C(dm,dm)
f = pm.rmv_normal_cov(np.sum([cv[name]*vals[name] for name in names],axis=0), C_eval) + np.random.normal(size=n_data+n_pred)*np.sqrt(V)
p = pm.flib.invlogit(f)
ns = 100
pos = pm.rbinomial(ns, p)
neg = ns - pos
print p
ra_data = np.rec.fromarrays((pos[:n_data], neg[:n_data], lon[:n_data], lat[:n_data]) + tuple([cv[name][:n_data] for name in names]), names=['pos','neg','lon','lat']+names)
pl.rec2csv(ra_data,'test_data.csv')
ra_pred = np.rec.fromarrays((pos[n_data:], neg[n_data:], lon[n_data:], lat[n_data:]) + tuple([cv[name][n_data:] for name in names]), names=['pos','neg','lon','lat']+names)
pl.rec2csv(ra_pred,'test_pred.csv')
os.system('infer cov_test test_db test_data.csv -t 10 -n 8 -i 100000')
# os.system('cov-test-predict test test_pred.csv 1000 100')
dims = 2.
beta_0 = 1.
nu_0 = dims
m_0 = np.zeros(dims)
W_0 = np.eye(dims)*1
N_points = 50
similarity = 1.
# Generate some data for two distributions
same = False
prior_deg_freedom = nu_0 +similarity# must be >= dims
prior_mu = m_0
prior_cov = W_0/similarity
prior_cov_wish = np.array(pymc.rwishart_cov(nu_0, W_0))#W_0#np.eye(dims)
true_mu1 = pymc.rmv_normal_cov(prior_mu,prior_cov)
true_cov1 = np.array(pymc.rwishart_cov(prior_deg_freedom, prior_cov_wish))
true_mu2 = pymc.rmv_normal_cov(prior_mu,prior_cov)
true_cov2 = np.array(pymc.rwishart_cov(prior_deg_freedom, prior_cov_wish))
if same:
true_mu2 = true_mu1
true_cov2 = true_cov1
#true_mu2 = true_mu1+0.1
obs1 = pymc.rmv_normal_cov(true_mu1, true_cov1, size = N_points)
obs2 = pymc.rmv_normal_cov(true_mu2, true_cov2, size = N_points)
all_obs = np.vstack((obs1,obs2))
all_labels = np.hstack((np.zeros(len(obs1)),np.ones(len(obs2))))
pylab.figure()
pylab.scatter(obs1[:,0],obs1[:,1])
# Set the parameters
dims = 2.
beta_0 = 1.
nu_0 = dims
m_0 = np.zeros(dims)
W_0 = np.eye(dims) * 1
N_points = 50
similarity = 1.
# Generate some data for two distributions
same = False
prior_deg_freedom = nu_0 + similarity # must be >= dims
prior_mu = m_0
prior_cov = W_0 / similarity
prior_cov_wish = np.array(pymc.rwishart_cov(nu_0, W_0)) #W_0#np.eye(dims)
true_mu1 = pymc.rmv_normal_cov(prior_mu, prior_cov)
true_cov1 = np.array(pymc.rwishart_cov(prior_deg_freedom, prior_cov_wish))
true_mu2 = pymc.rmv_normal_cov(prior_mu, prior_cov)
true_cov2 = np.array(pymc.rwishart_cov(prior_deg_freedom, prior_cov_wish))
if same:
true_mu2 = true_mu1
true_cov2 = true_cov1
#true_mu2 = true_mu1+0.1
obs1 = pymc.rmv_normal_cov(true_mu1, true_cov1, size=N_points)
obs2 = pymc.rmv_normal_cov(true_mu2, true_cov2, size=N_points)
all_obs = np.vstack((obs1, obs2))
all_labels = np.hstack((np.zeros(len(obs1)), np.ones(len(obs2))))
pylab.figure()
pylab.scatter(obs1[:, 0], obs1[:, 1])
def complex_hierarchical_data(n):
""" Generate data based on the much more complicated model
given in section 3.2.1::
y_ij ~ N(mu_j - exp(beta_j)t_ij - exp(gamma_j)t_ij^2, sigma_j^2)
gamma_j | sigma^2, xi, X_j ~ N(eta_0 + eta_1 X_j + eta_2 X_j^2, omega^2)
beta_j | gamma_j, sigma^2, xi, X_j ~ N(delta_beta_0 + delta_beta_1 X_j + delta_beta_2 X_j^2 + delta_beta_3 gamma_j, omega_beta^2)
mu_j | gamma_j, beta_j, sigma^2, xi, X_j ~ N(delta_mu_0 + delta_mu_1 X_j + delta_mu_2 X_j^2 + delta_mu_3 gamma_j + delta_mu_4 beta_j, omega_mu^2)
eta = (eta_0, eta_1, eta_2, log(omega))'
delta_beta = (delta_beta_0, delta_beta_1, delta_beta_2, delta_beta_3, log(omega_beta))'
delta_mu = (delta_mu_0, delta_mu_1, delta_mu_2, delta_mu_3, log(omega_mu))'
xi = (eta, delta_beta, delta_mu)
eta ~ MVNormal(M, C)
delta_beta, delta_mu ~ Normal(m, s)
Parameters
----------
n : list, len(n) = J, n[j] = num observations in group j
"""
J = len(n)
# covariate data, not entirely specified in paper
X = mc.rnormal(0, 0.1 ** -2, size=J)
t = [pl.arange(n[j]) for j in range(J)]
# hyper-priors, not specified in detail in paper
m = 0.0
s = 1.0
M = pl.zeros(4)
r = [[1, 0.57, 0.18, 0.56], [0.57, 1, 0.72, 0.16], [0.18, 0.72, 1, 0.14], [0.56, 0.16, 0.14, 1]]
eta = mc.rmv_normal_cov(M, r)
omega = 0.0001 # pl.exp(eta[-1])
delta_beta = mc.rnormal(m, s ** -2, size=5)
omega_beta = 0.0001 # pl.exp(delta_beta[-1])
delta_mu = mc.rnormal(m, s ** -2, size=5)
omega_mu = 0.0001 # pl.exp(delta_mu[-1])
gamma = mc.rnormal(eta[0] + eta[1] * X + eta[2] * X ** 2, omega ** -2.0)
beta = mc.rnormal(
delta_beta[0] + delta_beta[1] * X + delta_beta[2] * X ** 2 + delta_beta[3] * gamma, omega_beta ** -2
)
mu = mc.rnormal(
delta_mu[0] + delta_mu[1] * X + delta_mu[2] * X ** 2 + delta_mu[3] * gamma + delta_mu[4] * beta, omega_mu ** -2
)
# stochastic error, not specified in paper
sigma = 0.01 * pl.ones(J)
y = [mc.rnormal(mu[j] - pl.exp(beta[j]) * t[j] - pl.exp(gamma[j]) * t[j] ** 2, sigma[j] ** -2) for j in range(J)]
eta_cross_eta = [
eta[0] * eta[1],
eta[0] * eta[2],
eta[0] * eta[3],
eta[1] * eta[2],
eta[1] * eta[2],
eta[2] * eta[3],
]
return vars()
[ \rho\sigma_2\sigma_1 \sigma_2\sigma_2 ]
Then, knowing the covariance matrix C and given the random samples,
we want to estimate the posterior distribution of mu. Since the prior
for mu is uniform, the mean posterior distribution is simply a bivariate
normal with the same correlation coefficient rho, but with variances
divided by sqrt(N), where N is the number of samples drawn.
We can check that the sampler works correctly by making sure that
after a while, the covariance matrix of the samples for mu tend to C/N.
"""
N=50
mu = np.array([-2., 3.])
C = np.array([[1, .8*np.sqrt(2)], [.8*np.sqrt(2), 2.]])
r = pymc.rmv_normal_cov(mu, C, size=50)
@pymc.stoch
def mean(value=np.array([0., 0.])):
"""The mean of the samples (mu). """
return 0.
obs = pymc.MvNormalCov('obs', mean, C, value=r, observed=True)
class TestAM(TestCase):
def test_convergence(self):
S = pymc.MCMC([mean, obs])
S.use_step_method(pymc.AdaptiveMetropolis, mean, delay=200)
S.sample(6000, burn=1000)
Cs = np.cov(S.trace('mean')[:].T)
