def test_gaussian_em():
"""Test the Gaussian EM on a small generated dataset"""
fname = "gmm-3-10-0.7.npz"
gmm = GaussianMixtureModel.generate(fname, 3, 3)
k, d, M, S, w = gmm.k, gmm.d, gmm.means, gmm.sigmas, gmm.weights
N, n = 1e6, 1e5
X = gmm.sample(N, n)
algo = GaussianMixtureEM(k, d)
def report(i, O_, lhood):
M_, _, _ = O_
lhood, Z, O_ = algo.run(X, None, report)
M_, S_, w_ = O_
M_ = closest_permuted_matrix(M, M_)
w_ = closest_permuted_vector(w, w_)
print w, w_
print norm(M - M_) / norm(M)
print abs(S - S_).max()
print norm(w - w_)
assert (norm(M - M_) / norm(M) < 1e-1)
assert (abs(S - S_) < 1).all()
assert (norm(w - w_) < 1e-2)
def test_univarate(n=3, d=20, e=20):
# for sigma in [0, 1e-4, 1e-3, 1e-2]:
for _ in xrange(100):
sigma = 0
R, I, V = generate_univariate_problem(n, d, e)
print "V", V
I = add_noise(I, sigma)
for eps in [1e-7, 1e-6, 1e-5, 1e-4]:
try:
V_ = BB.BorderBasisFactory(eps + sigma).generate(R, I).zeros()
# Find the maximal matching between V and V_
V_ = array(V_).flatten()
print "V_", V_
print "e", eps
break
except AssertionError:
continue
else:
print "could not find 0"
return R, I, V
if len(V) == len(V_):
V_ = closest_permuted_vector(V, V_)
print "diff", (V - V_)
if max(abs(V - V_)) > max(1e-3, sigma):
print "error too large!"
return R, I, V
else:
print "incorrect lengths!"
return R, I, V
def get_univariate_stats(n=3, d=20, e=20, sigma=0.0, tries=100):
status, epses, err = [], [], []
for _ in xrange(tries):
R, I, V = generate_univariate_problem(n, d, e)
I = add_noise(I, sigma)
for eps in [1e-7, 1e-6, 1e-5, 1e-4, 1e-3, 1e-2, 1e-1]:
try:
V_ = BB.BorderBasisFactory(sigma + eps).generate(R, I).zeros()
V_ = array(V_).flatten()
epses.append(eps)
break
except AssertionError:
continue
else:
status.append(1)
epses.append(1)
err.append(1)
continue
if len(V_) < len(V):
status.append(2)
# Pad V_.
V_ = np.hstack((V_, np.zeros(len(V) - len(V_))))
elif len(V_) > len(V):
status.append(3)
# Pad V_.
V = np.hstack((V, np.zeros(len(V_) - len(V))))
else:
status.append(4)
V_ = closest_permuted_vector(V, V_)
err.append(norm(V - V_))
return status, epses, err
def test_gaussian_em():
"""Test the Gaussian EM on a small generated dataset"""
fname = "gmm-3-10-0.7.npz"
gmm = GaussianMixtureModel.generate( fname, 3, 3 )
k, d, M, S, w = gmm.k, gmm.d, gmm.means, gmm.sigmas, gmm.weights
N, n = 1e6, 1e5
X = gmm.sample( N, n )
algo = GaussianMixtureEM(k, d)
def report( i, O_, lhood ):
M_, _, _ = O_
lhood, Z, O_ = algo.run( X, None, report )
M_, S_, w_ = O_
M_ = closest_permuted_matrix( M, M_ )
w_ = closest_permuted_vector( w, w_ )
print w, w_
print norm( M - M_ )/norm(M)
print abs(S - S_).max()
print norm( w - w_ )
assert( norm( M - M_ )/norm(M) < 1e-1 )
assert (abs(S - S_) < 1 ).all()
assert( norm( w - w_ ) < 1e-2 )
def test_univarate(n=3, d=20, e=20):
#for sigma in [0, 1e-4, 1e-3, 1e-2]:
for _ in xrange(100):
sigma = 0
R, I, V = generate_univariate_problem(n, d, e)
print "V", V
I = add_noise(I, sigma)
for eps in [1e-7, 1e-6, 1e-5, 1e-4]:
try:
V_ = BB.BorderBasisFactory(eps + sigma).generate(R, I).zeros()
# Find the maximal matching between V and V_
V_ = array(V_).flatten()
print "V_", V_
print 'e', eps
break
except AssertionError:
continue
else:
print "could not find 0"
return R, I, V
if len(V) == len(V_):
V_ = closest_permuted_vector(V, V_)
print "diff", (V - V_)
if max(abs(V - V_)) > max(1e-3, sigma):
print "error too large!"
return R, I, V
else:
print "incorrect lengths!"
return R, I, V
def get_univariate_stats(n=3, d=20, e=20, sigma=0., tries=100):
status, epses, err = [], [], []
for _ in xrange(tries):
R, I, V = generate_univariate_problem(n, d, e)
I = add_noise(I, sigma)
for eps in [1e-7, 1e-6, 1e-5, 1e-4, 1e-3, 1e-2, 1e-1]:
try:
V_ = BB.BorderBasisFactory(sigma + eps).generate(R, I).zeros()
V_ = array(V_).flatten()
epses.append(eps)
break
except AssertionError:
continue
else:
status.append(1)
epses.append(1)
err.append(1)
continue
if len(V_) < len(V):
status.append(2)
# Pad V_.
V_ = np.hstack((V_, np.zeros(len(V) - len(V_))))
elif len(V_) > len(V):
status.append(3)
# Pad V_.
V = np.hstack((V, np.zeros(len(V_) - len(V))))
else:
status.append(4)
V_ = closest_permuted_vector(V, V_)
err.append(norm(V - V_))
return status, epses, err