def testParameterEstimation(self):
d = 3
mu = numpy.ones(d)
sigma = numpy.eye(d)
pdf = MultivariateGaussian(mu, sigma)
samples = pdf.random(100000)
pdf.estimate(samples)
for i in range(d):
self.assertAlmostEqual(pdf.mu[i], mu[i], delta=1e-1)
for j in range(d):
self.assertAlmostEqual(pdf.sigma[i, j],
sigma[i, j],
delta=1e-1)
d = 3
mu = numpy.array([0., 1., 2.])
sigma = numpy.random.random((d, d))
sigma = numpy.dot(sigma, sigma.T)
pdf = MultivariateGaussian(mu, sigma)
samples = pdf.random(1000000)
pdf.estimate(samples)
for i in range(d):
self.assertAlmostEqual(pdf.mu[i], mu[i], delta=1e-1)
for j in range(d):
self.assertAlmostEqual(pdf.sigma[i, j],
sigma[i, j],
delta=1e-1)
def testParameters(self):
pdf = MultivariateGaussian(numpy.zeros(3), numpy.eye(3))
d = 2
pdf.mu = numpy.zeros(d)
pdf.sigma = numpy.eye(d)
self.assertEqual(list(pdf.mu), [0, 0])
self.assertEqual(pdf.sigma.tolist(), numpy.eye(d).tolist())
def testParameters(self):
pdf = MultivariateGaussian(numpy.zeros(3), numpy.eye(3))
d = 2
pdf.mu = numpy.zeros(d)
pdf.sigma = numpy.eye(d)
self.assertEqual(list(pdf.mu), [0, 0])
self.assertEqual(pdf.sigma.tolist(), numpy.eye(d).tolist())
def testParameterEstimation(self):
d = 3
mu = numpy.ones(d)
sigma = numpy.eye(d)
pdf = MultivariateGaussian(mu, sigma)
samples = pdf.random(100000)
pdf.estimate(samples)
for i in range(d):
self.assertAlmostEqual(pdf.mu[i], mu[i], delta=1e-1)
for j in range(d):
self.assertAlmostEqual(pdf.sigma[i, j], sigma[i, j], delta=1e-1)
d = 3
mu = numpy.array([0., 1., 2.])
sigma = numpy.random.random((d, d))
sigma = numpy.dot(sigma, sigma.T)
pdf = MultivariateGaussian(mu, sigma)
samples = pdf.random(1000000)
pdf.estimate(samples)
for i in range(d):
self.assertAlmostEqual(pdf.mu[i], mu[i], delta=1e-1)
for j in range(d):
self.assertAlmostEqual(pdf.sigma[i, j], sigma[i, j], delta=1e-1)
def testLogProb(self):
mg = MultivariateGaussian(numpy.zeros(2), numpy.eye(2))
self.assertAlmostEqual(mg(numpy.zeros(2)),
1. / (2 * numpy.pi),
delta=1e-1)
self.assertAlmostEqual(mg(numpy.ones(2)),
1. / (2 * numpy.pi) * numpy.exp(-0.5),
delta=1e-1)
coords = StructureParser(prot1 + '.pdb').parse().get_coordinates(['CA']) / 4.0
coords2 = StructureParser(prot2 + '.pdb').parse().get_coordinates(['CA']) / 4.0
## size of fake ensemble
ensemble_size = 1000
## sigma of Gausian noise added to positions to generate random ensemble
sigma = 0.05
## cutoff from which to determine contact frequencies
cutoff = 1.8
n_beads = len(coords)
if False:
## create data by adding Gaussian noise
g = MultivariateGaussian(mu=coords.ravel(), sigma=sigma * numpy.eye(n_beads*3))
ensemble1 = g.random(size=ensemble_size).reshape(ensemble_size, n_beads, 3)
g = MultivariateGaussian(mu=coords2.ravel(), sigma=sigma * numpy.eye(n_beads*3))
ensemble2 = g.random(size=ensemble_size).reshape(ensemble_size, n_beads, 3)
dms1 = numpy.array(map(distance_matrix, ensemble1))
dms2 = numpy.array(map(distance_matrix, ensemble2))
frequencies1 = numpy.sum(dms1 < cutoff, 0)
frequencies2 = numpy.sum(dms2 < cutoff, 0)
summed_frequencies = frequencies1 + frequencies2
if False:
## create Poisson-distributed data
dm1 = distance_matrix(coords)
dm2 = distance_matrix(coords2)
cs1 = ensemble_size * (dm1 < cutoff)
cs2 = ensemble_size * (dm2 < cutoff)
