def test_log_post_pred():
# Data generated with np.random.seed(2); np.random.rand(11, 4)
X = np.array([[0.4359949, 0.02592623, 0.54966248, 0.43532239],
[0.4203678, 0.33033482, 0.20464863, 0.61927097],
[0.29965467, 0.26682728, 0.62113383, 0.52914209],
[0.13457995, 0.51357812, 0.18443987, 0.78533515],
[0.85397529, 0.49423684, 0.84656149, 0.07964548],
[0.50524609, 0.0652865, 0.42812233, 0.09653092],
[0.12715997, 0.59674531, 0.226012, 0.10694568],
[0.22030621, 0.34982629, 0.46778748, 0.20174323],
[0.64040673, 0.48306984, 0.50523672, 0.38689265],
[0.79363745, 0.58000418, 0.1622986, 0.70075235],
[0.96455108, 0.50000836, 0.88952006, 0.34161365]])
N, D = X.shape
# Setup densities
m_0 = X.mean(axis=0)
k_0 = 0.05
v_0 = D + 10
S_0 = 0.5 * np.ones(D)
prior = NIW(m_0, k_0, v_0, S_0)
gmm = GaussianComponentsDiag(X, prior, [0, 0, 0, 1, 0, 1, 3, 4, 3, 2, -1])
expected_log_post_pred = log_post_pred_unvectorized(gmm, 10)
npt.assert_almost_equal(gmm.log_post_pred(10), expected_log_post_pred)
def test_log_marg_k():
np.random.seed(1)
# Generate data
D = 10
N_1 = 10
X_1 = 5 * np.random.rand(N_1, D) - 1
# Prior
m_0 = 5 * np.random.rand(D) - 2
k_0 = np.random.randint(15)
v_0 = D + np.random.randint(5)
S_0 = 2 * np.random.rand(D) + 3
prior = NIW(m_0=m_0, k_0=k_0, v_0=v_0, S_0=S_0)
# Setup GMM
assignments = np.concatenate([np.zeros(N_1)])
gmm = GaussianComponentsDiag(X_1, prior, assignments=assignments)
# Calculate marginal for component by hand
k_N = k_0 + N_1
v_N = v_0 + N_1
m_N = (k_0 * m_0 + N_1 * X_1.mean(axis=0)) / k_N
S_N = S_0 + np.square(X_1).sum(
axis=0) + k_0 * np.square(m_0) - k_N * np.square(m_N)
var = S_N * (k_N + 1) / (k_N * v_N)
expected_log_marg = (-N_1 * D / 2. * math.log(np.pi) +
D / 2. * math.log(k_0) - D / 2. * math.log(k_N) +
v_0 / 2. * np.log(S_0).sum() -
v_N / 2. * np.log(S_N).sum() + D *
(gammaln(v_N / 2.) - gammaln(v_0 / 2.)))
npt.assert_almost_equal(gmm.log_marg_k(0), expected_log_marg)
def test_3component_with_delete_post_pred_k():
np.random.seed(1)
# Generate data
D = 10
N_1 = 10
N_2 = 5
N_3 = 5
X = 5 * np.random.rand(N_1 + N_2 + N_3, D) - 1
X_1 = X[:N_1]
X_2 = X[N_1:N_1 + N_2]
X_3 = X[N_1 + N_2:]
# Prior
m_0 = 5 * np.random.rand(D) - 2
k_0 = np.random.randint(15)
v_0 = D + np.random.randint(5)
S_0 = 2 * np.random.rand(D) + 3
prior = NIW(m_0=m_0, k_0=k_0, v_0=v_0, S_0=S_0)
# Setup GMM
assignments = np.concatenate(
[np.zeros(N_1), np.ones(N_2), 2 * np.ones(N_3)])
gmm = GaussianComponentsDiag(X, prior, assignments=assignments)
# Remove everything from component 2
for i in range(N_1, N_1 + N_2):
gmm.del_item(i)
# Calculate posterior for first component by hand
x_1 = X_1[0]
k_N = k_0 + N_1
v_N = v_0 + N_1
m_N = (k_0 * m_0 + N_1 * X_1.mean(axis=0)) / k_N
S_N = S_0 + np.square(X_1).sum(
axis=0) + k_0 * np.square(m_0) - k_N * np.square(m_N)
var = S_N * (k_N + 1) / (k_N * v_N)
expected_posterior = np.sum([
students_t(x_1[i], m_N[i], S_N[i] * (k_N + 1) / (k_N * v_N), v_N)
for i in range(len(x_1))
])
npt.assert_almost_equal(gmm.log_post_pred_k(0, 0), expected_posterior)
# Calculate posterior for second component by hand
x_1 = X_3[0]
k_N = k_0 + N_3
v_N = v_0 + N_3
m_N = (k_0 * m_0 + N_3 * X_3.mean(axis=0)) / k_N
S_N = S_0 + np.square(X_3).sum(
axis=0) + k_0 * np.square(m_0) - k_N * np.square(m_N)
var = S_N * (k_N + 1) / (k_N * v_N)
expected_posterior = np.sum([
students_t(x_1[i], m_N[i], S_N[i] * (k_N + 1) / (k_N * v_N), v_N)
for i in range(len(x_1))
])
npt.assert_almost_equal(gmm.log_post_pred_k(N_1 + N_2, 1),
expected_posterior)
def test_log_post_pred_k():
np.random.seed(1)
# Prior
D = 10
m_0 = 5 * np.random.rand(D) - 2
k_0 = np.random.randint(15)
v_0 = D + np.random.randint(5)
S_0 = 2 * np.random.rand(D) + 3
prior = NIW(m_0=m_0, k_0=k_0, v_0=v_0, S_0=S_0)
# Data
N = 12
X = 5 * np.random.rand(N, D) - 1
# Setup GMM
gmm = GaussianComponentsDiag(X, prior)
for i in range(N):
gmm.add_item(i, 0)
# Calculate posterior by hand
x = X[0]
k_N = k_0 + N
v_N = v_0 + N
m_N = (k_0 * m_0 + N * X[:N].mean(axis=0)) / k_N
S_N = S_0 + np.square(
X[:N]).sum(axis=0) + k_0 * np.square(m_0) - k_N * np.square(m_N)
var = S_N * (k_N + 1) / (k_N * v_N)
expected_posterior = np.sum([
students_t(x[i], m_N[i], S_N[i] * (k_N + 1) / (k_N * v_N), v_N)
for i in range(len(x))
])
npt.assert_almost_equal(gmm.log_post_pred_k(0, 0), expected_posterior)
def test_2component_post_pred_k():
np.random.seed(1)
# Generate data
D = 10
N_1 = 10
N_2 = 5
X = 5 * np.random.rand(N_1 + N_2, D) - 1
X_1 = X[:N_1]
X_2 = X[N_1:]
# Prior
m_0 = 5 * np.random.rand(D) - 2
k_0 = np.random.randint(15)
v_0 = D + np.random.randint(5)
S_0 = 2 * np.random.rand(D) + 3
prior = NIW(m_0=m_0, k_0=k_0, v_0=v_0, S_0=S_0)
# Setup GMM
assignments = np.concatenate([np.zeros(N_1), np.ones(N_2)])
gmm = GaussianComponentsDiag(X, prior, assignments=assignments)
# Remove one item (as additional check)
gmm.del_item(N_1 + N_2 - 1)
X_2 = X_2[:-1]
N_2 -= 1
# Calculate posterior for first component by hand
x_1 = X_1[0]
k_N = k_0 + N_1
v_N = v_0 + N_1
m_N = (k_0 * m_0 + N_1 * X_1.mean(axis=0)) / k_N
S_N = S_0 + np.square(X_1).sum(
axis=0) + k_0 * np.square(m_0) - k_N * np.square(m_N)
var = S_N * (k_N + 1) / (k_N * v_N)
expected_posterior = np.sum([
students_t(x_1[i], m_N[i], S_N[i] * (k_N + 1) / (k_N * v_N), v_N)
for i in range(len(x_1))
])
npt.assert_almost_equal(gmm.log_post_pred_k(0, 0), expected_posterior)
# Calculate posterior for second component by hand
x_1 = X_2[0]
k_N = k_0 + N_2
v_N = v_0 + N_2
m_N = (k_0 * m_0 + N_2 * X_2.mean(axis=0)) / k_N
S_N = S_0 + np.square(X_2).sum(
axis=0) + k_0 * np.square(m_0) - k_N * np.square(m_N)
var = S_N * (k_N + 1) / (k_N * v_N)
expected_posterior = np.sum([
students_t(x_1[i], m_N[i], S_N[i] * (k_N + 1) / (k_N * v_N), v_N)
for i in range(len(x_1))
])
npt.assert_almost_equal(gmm.log_post_pred_k(N_1, 1), expected_posterior)
def main():
# Load data
X = pickle.load(open(data_fn, "rb"))
N, D = X.shape
# Model parameters
alpha = 1.
K = 4 # number of components
mu_scale = 3.0
covar_scale = 1.0
# Sampling parameters
n_runs = 2
n_iter = 12
# Intialize prior
m_0 = np.zeros(D)
k_0 = covar_scale**2 / mu_scale**2
v_0 = D + 3
S_0 = covar_scale**2 * v_0 * np.eye(D)
prior = NIW(m_0, k_0, v_0, S_0)
# Initialize component assignment: this is not random for testing purposes
z = np.array([i * np.ones(N / K) for i in range(K)],
dtype=np.int).flatten()
# Setup FBGMM
fbgmm = FBGMM(X, prior, alpha, K, assignments=z)
print("Initial log marginal prob:", fbgmm.log_marg())
# Perform several Gibbs sampling runs and average the log marginals
log_margs = np.zeros(n_iter)
for j in range(n_runs):
# Perform Gibbs sampling
record = fbgmm.gibbs_sample(n_iter)
log_margs += record["log_marg"]
log_margs /= n_runs
# Plot results
fig = plt.figure()
ax = fig.add_subplot(111)
plot_mixture_model(ax, fbgmm)
for k in range(fbgmm.components.K):
mu, sigma = fbgmm.components.rand_k(k)
plot_ellipse(ax, mu, sigma)
# Plot log probability
plt.figure()
plt.plot(list(range(n_iter)), log_margs)
plt.xlabel("Iterations")
plt.ylabel("Log prob")
plt.show()
def gmm(X,
K=4,
n_iter=100,
alpha=1.0,
mu_scale=4.0,
var_scale=0.5,
covar_scale=0.7,
posterior_predictive_check=False):
N, D = X.shape
# Initialize prior
m_0 = np.zeros(D)
k_0 = covar_scale**2 / mu_scale**2
v_0 = D + 3
S_0 = covar_scale**2 * v_0 * np.ones(D)
prior = NIW(m_0, k_0, v_0, S_0)
# Setup FBGMM
fbgmm = FBGMM(X, prior, alpha, K, "rand")
# Perform Gibbs sampling
record = fbgmm.gibbs_sample(n_iter)
K = fbgmm.components.K
mus = np.zeros(shape=(K, D))
covars = [np.zeros((D, D)) for i in range(0, K)]
for k in range(fbgmm.components.K):
mu, var = fbgmm.components.rand_k(k)
mus[k, :] = mu
covars[k] = np.diag(var)
# Generate new points for posterior predictive check
# Generate the same number of points as N
if posterior_predictive_check:
np.random.seed(1)
rstate = 1
alphas = (alpha / K) + fbgmm.components.counts
pis = dirichlet.rvs(alphas, random_state=rstate)[0]
Z = np.zeros(N, dtype=np.uint32)
X = np.zeros((N, D))
for n in range(N):
Z[n] = np.floor(np.argmax(multinomial(1, pis)))
X[n] = multivariate_normal.rvs(mean=mus[Z[n]], cov=covars[Z[n]])
return fbgmm.components.assignments, mus, (X, Z)
return fbgmm.components.assignments, mus
def test_log_post_pred_k():
# Setup densities
prior = NIW(m_0=np.array([0.0, 0.0]), k_0=2., v_0=5., S_0=5. * np.eye(2))
gmm = GaussianComponents(np.array([[1.2, 0.9], [-0.1, 0.8], [0.5, 0.4]]),
prior)
# Add data vectors to a single component
gmm.add_item(0, 0)
gmm.add_item(1, 0)
# Calculate log predictave
lp = gmm.log_post_pred_k(2, 0)
lp_expected = -2.07325364088
npt.assert_almost_equal(lp, lp_expected)
def main():
# Data parameters
D = 2 # dimensions
N = 100 # number of points to generate
K_true = 4 # the true number of components
# Model parameters
alpha = 1.
K = 3 # initial number of components
n_iter = 20
# Generate data
mu_scale = 4.0
covar_scale = 0.7
z_true = np.random.randint(0, K_true, N)
mu = np.random.randn(D, K_true) * mu_scale
X = mu[:, z_true] + np.random.randn(D, N) * covar_scale
X = X.T
# Intialize prior
m_0 = np.zeros(D)
k_0 = covar_scale**2 / mu_scale**2
v_0 = D + 3
S_0 = covar_scale**2 * v_0 * np.ones(D)
prior = NIW(m_0, k_0, v_0, S_0)
# Setup IGMM
igmm = IGMM(X,
prior,
alpha,
assignments="rand",
K=K,
covariance_type="diag")
# igmm = IGMM(X, prior, alpha, assignments="one-by-one", K=K)
# Perform Gibbs sampling
record = igmm.gibbs_sample(n_iter)
# Plot results
fig = plt.figure()
ax = fig.add_subplot(111)
plot_mixture_model(ax, igmm)
for k in xrange(igmm.components.K):
mu, sigma = igmm.components.rand_k(k)
plot_ellipse(ax, mu, np.diag(sigma))
plt.show()
def test_sampling_2d_assignments():
random.seed(1)
np.random.seed(1)
# Data parameters
D = 2 # dimensions
N = 100 # number of points to generate
K_true = 4 # the true number of components
# Model parameters
alpha = 1.
K = 3 # number of components
n_iter = 10
# Generate data
mu_scale = 4.0
covar_scale = 0.7
z_true = np.random.randint(0, K_true, N)
mu = np.random.randn(D, K_true)*mu_scale
X = mu[:, z_true] + np.random.randn(D, N)*covar_scale
X = X.T
# Intialize prior
m_0 = np.zeros(D)
k_0 = covar_scale**2/mu_scale**2
v_0 = D + 3
S_0 = covar_scale**2*v_0*np.eye(D)
prior = NIW(m_0, k_0, v_0, S_0)
# Setup FBGMM
fbgmm = FBGMM(X, prior, alpha, K, "rand")
# Perform Gibbs sampling
record = fbgmm.gibbs_sample(n_iter)
assignments_expected = np.array([
0, 2, 0, 0, 2, 0, 2, 2, 2, 0, 0, 0, 0, 2, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0,
2, 0, 1, 0, 2, 1, 1, 0, 2, 2, 0, 0, 2, 1, 0, 1, 0, 0, 0, 2, 2, 0, 1, 0,
0, 0, 0, 0, 0, 0, 0, 2, 2, 1, 0, 0, 1, 2, 2, 1, 0, 0, 0, 2, 0, 0, 0, 2,
0, 1, 0, 0, 0, 2, 2, 1, 2, 0, 0, 0, 2, 1, 2, 2, 1, 0, 0, 1, 0, 2, 2, 1,
2, 0, 0, 2
])
assignments = fbgmm.components.assignments
npt.assert_array_equal(assignments, assignments_expected)
def test_del_item():
np.random.seed(1)
# Prior
D = 10
m_0 = 5 * np.random.rand(D) - 2
k_0 = np.random.randint(15)
v_0 = D + np.random.randint(5)
S_0 = 2 * np.random.rand(D) + 3
prior = NIW(m_0=m_0, k_0=k_0, v_0=v_0, S_0=S_0)
# Data
N = 12
X = 5 * np.random.rand(N, D) - 1
# Setup GMM
gmm = GaussianComponentsDiag(X, prior)
for i in range(N):
gmm.add_item(i, 0)
# Remove 5 random items
del_items = set(np.random.randint(1, N, size=5))
for i in del_items:
gmm.del_item(i)
indices = list(set(range(N)).difference(del_items))
# Calculate posterior by hand
X = X[indices]
N, _ = X.shape
x = X[0]
k_N = k_0 + N
v_N = v_0 + N
m_N = (k_0 * m_0 + N * X[:N].mean(axis=0)) / k_N
S_N = S_0 + np.square(
X[:N]).sum(axis=0) + k_0 * np.square(m_0) - k_N * np.square(m_N)
var = S_N * (k_N + 1) / (k_N * v_N)
expected_posterior = np.sum([
students_t(x[i], m_N[i], S_N[i] * (k_N + 1) / (k_N * v_N), v_N)
for i in range(len(x))
])
npt.assert_almost_equal(gmm.log_post_pred_k(0, 0), expected_posterior)
def test_log_prior_3d():
# Data
X = np.array([[-0.3406, -0.0593, -0.0686]])
N, D = X.shape
# Setup densities
m_0 = np.zeros(D)
k_0 = 0.05
v_0 = D + 1
S_0 = 0.001 * np.eye(D)
prior = NIW(m_0, k_0, v_0, S_0)
gmm = GaussianComponents(X, prior)
# Calculate log predictave under prior alone
lp = gmm.log_prior(0)
lp_expected = -0.472067277015
npt.assert_almost_equal(lp, lp_expected)
def test_sampling_2d_log_marg_deleted_components():
random.seed(1)
np.random.seed(1)
# Data parameters
D = 2 # dimensions
N = 10 # number of points to generate
K_true = 4 # the true number of components
# Model parameters
alpha = 1.
K = 6 # number of components
n_iter = 1
# Generate data
mu_scale = 4.0
covar_scale = 0.7
z_true = np.random.randint(0, K_true, N)
mu = np.random.randn(D, K_true)*mu_scale
X = mu[:, z_true] + np.random.randn(D, N)*covar_scale
X = X.T
# Intialize prior
m_0 = np.zeros(D)
k_0 = covar_scale**2/mu_scale**2
v_0 = D + 3
S_0 = covar_scale**2*v_0*np.eye(D)
prior = NIW(m_0, k_0, v_0, S_0)
# Setup FBGMM
fbgmm = FBGMM(X, prior, alpha, K, "rand")
# Perform Gibbs sampling
record = fbgmm.gibbs_sample(n_iter)
expected_log_marg = -60.1448630929
log_marg = fbgmm.log_marg()
print(fbgmm.components.assignments)
npt.assert_almost_equal(log_marg, expected_log_marg)
def main():
# Data parameters
D = 2 # dimensions
N = 100 # number of points to generate
K_true = 4 # the true number of components
# Model parameters
alpha = 1.
K = 4 # number of components
n_iter = 20
# Generate data
mu_scale = 4.0
covar_scale = 0.7
z_true = np.random.randint(0, K_true, N)
mu = np.random.randn(D, K_true) * mu_scale
X = mu[:, z_true] + np.random.randn(D, N) * covar_scale
X = X.T
# Intialize prior
m_0 = np.zeros(D)
k_0 = covar_scale**2 / mu_scale**2
v_0 = D + 3
S_0 = covar_scale**2 * v_0 * np.eye(D)
prior = NIW(m_0, k_0, v_0, S_0)
# Setup FBGMM
fbgmm = FBGMM(X, prior, alpha, K, "rand")
# Perform Gibbs sampling
record = fbgmm.gibbs_sample(n_iter)
# Plot results
fig = plt.figure()
ax = fig.add_subplot(111)
plot_mixture_model(ax, fbgmm)
for k in range(fbgmm.components.K):
# mu, sigma = fbgmm.components.map(k)
mu, sigma = fbgmm.components.rand_k(k)
plot_ellipse(ax, mu, sigma)
plt.show()
def test_sampling_2d_log_marg_deleted_components():
random.seed(2)
np.random.seed(2)
# Data parameters
D = 2 # dimensions
N = 5 # number of points to generate
K_true = 4 # the true number of components
# Model parameters
alpha = 1.
K = 3 # initial number of components
n_iter = 1
# Generate data
mu_scale = 4.0
covar_scale = 0.7
z_true = np.random.randint(0, K_true, N)
mu = np.random.randn(D, K_true) * mu_scale
X = mu[:, z_true] + np.random.randn(D, N) * covar_scale
X = X.T
print(X)
# Intialize prior
m_0 = np.zeros(D)
k_0 = covar_scale**2 / mu_scale**2
v_0 = D + 3
S_0 = covar_scale**2 * v_0 * np.eye(D)
prior = NIW(m_0, k_0, v_0, S_0)
# Setup IGMM
igmm = IGMM(X, prior, alpha, assignments="each-in-own")
# Perform Gibbs sampling
record = igmm.gibbs_sample(n_iter)
print(igmm.components.assignments)
expected_log_marg = -30.771535771
log_marg = igmm.log_marg()
npt.assert_almost_equal(log_marg, expected_log_marg)
def test_map():
# Setup densities
prior = NIW(m_0=np.array([0.0, 0.0]),
k_0=2.0,
v_0=5.0,
S_0=5.0 * np.eye(2))
gmm = GaussianComponents(np.array([[1.2, 0.9], [-0.1, 0.8]]), prior)
gmm.add_item(0, 0)
gmm.add_item(1, 0)
mu_expected = np.array([0.275, 0.425])
sigma_expected = np.array([[0.55886364, 0.04840909],
[0.04840909, 0.52068182]])
# Calculate the posterior MAP of the parameters
mu, sigma = gmm.map(0)
npt.assert_almost_equal(mu, mu_expected)
npt.assert_almost_equal(sigma, sigma_expected)
def test_sampling_2d_assignments_deleted_components():
random.seed(1)
np.random.seed(1)
# Data parameters
D = 2 # dimensions
N = 20 # number of points to generate
K_true = 4 # the true number of components
# Model parameters
alpha = 1.
K = 3 # initial number of components
n_iter = 1
# Generate data
mu_scale = 4.0
covar_scale = 0.7
z_true = np.random.randint(0, K_true, N)
mu = np.random.randn(D, K_true) * mu_scale
X = mu[:, z_true] + np.random.randn(D, N) * covar_scale
X = X.T
# Intialize prior
m_0 = np.zeros(D)
k_0 = covar_scale**2 / mu_scale**2
v_0 = 5
S_0 = covar_scale**2 * v_0 * np.eye(D)
prior = NIW(m_0, k_0, v_0, S_0)
# Setup IGMM
igmm = IGMM(X, prior, alpha, assignments="each-in-own")
# Perform Gibbs sampling
record = igmm.gibbs_sample(n_iter)
assignments_expected = np.array(
[5, 2, 4, 3, 2, 7, 2, 7, 1, 0, 4, 6, 4, 1, 6, 4, 1, 7, 1, 0])
assignments = igmm.components.assignments
npt.assert_array_equal(assignments, assignments_expected)
def test_log_marg_k():
# Data
X = np.array([[-0.3406, -0.3593, -0.0686], [-0.3381, 0.2993, 0.925],
[-0.5, -0.101, 0.75]])
N, D = X.shape
# Setup densities
m_0 = np.zeros(D)
k_0 = 0.05
v_0 = D + 3
S_0 = 0.5 * np.eye(D)
prior = NIW(m_0, k_0, v_0, S_0)
gmm = GaussianComponents(X, prior, [0, 0, 0])
log_marg_expected = -8.42365141729
# Calculate log marginal of data
log_marg = gmm.log_marg_k(0)
npt.assert_almost_equal(log_marg, log_marg_expected)
def test_sampling_2d_log_marg():
random.seed(1)
np.random.seed(1)
# Data parameters
D = 2 # dimensions
N = 100 # number of points to generate
K_true = 4 # the true number of components
# Model parameters
alpha = 1.
K = 3 # initial number of components
n_iter = 10
# Generate data
mu_scale = 4.0
covar_scale = 0.7
z_true = np.random.randint(0, K_true, N)
mu = np.random.randn(D, K_true) * mu_scale
X = mu[:, z_true] + np.random.randn(D, N) * covar_scale
X = X.T
# Intialize prior
m_0 = np.zeros(D)
k_0 = covar_scale**2 / mu_scale**2
v_0 = 5
S_0 = covar_scale**2 * v_0 * np.eye(D)
prior = NIW(m_0, k_0, v_0, S_0)
# Setup IGMM
igmm = IGMM(X, prior, alpha, K=K)
# Perform Gibbs sampling
record = igmm.gibbs_sample(n_iter)
expected_log_marg = -411.811711231
log_marg = igmm.log_marg()
npt.assert_almost_equal(log_marg, expected_log_marg)
def test_log_prod_students_t():
np.random.seed(1)
# Prior
D = 10
m_0 = 5 * np.random.rand(D) - 2
k_0 = np.random.randint(15)
v_0 = D + np.random.randint(5)
S_0 = 2 * np.random.rand(D) + 3
prior = NIW(m_0=m_0, k_0=k_0, v_0=v_0, S_0=S_0)
# GMM we will use to access `_log_prod_students_t`
x = 3 * np.random.rand(D) + 4
gmm = GaussianComponentsDiag(np.array([x]), prior)
expected_prior = np.sum([
students_t(x[i], m_0[i], S_0[i] * (k_0 + 1) / (k_0 * v_0), v_0)
for i in range(len(x))
])
npt.assert_almost_equal(gmm.log_prior(0), expected_prior)