def check_probability(self, model):
samples_count = 100
components_count = model.GetNumberOfPrincipalComponents()
for _ in range(0, samples_count):
coeffs = randn(components_count)
s = model.DrawSample(coeffs)
p = model.ComputeProbability(s)
# as the distribution we are looking for is just a standard mv normal, all the
# components are independent. We can thus use a 1d normal distribution to compute the
# probability of the full pdf.
p_with_spicy = 1
for c in coeffs:
p_with_spicy *= normdist(0, 1).pdf(c)
self.assertTrue(p, p_with_spicy)
def testProbabilityOfDatasetPlausibility(self):
num_samples = 100
nComps = self.model.GetNumberOfPrincipalComponents()
p_mean = self.model.ComputeProbability(self.model.DrawMean())
for i in xrange(0, num_samples):
coeffs = randn(nComps)
s = self.model.DrawSample(coeffs)
p = self.model.ComputeProbability(s)
# as the distribution we are looking for is just a standard mv normal, all the
# components are independent. We can thus use a 1d normal distribution to compute the
# probability of the full pdf.
pScipy = 1
for c in coeffs:
pScipy *= normdist(0, 1).pdf(c)
self.assertTrue(p, pScipy)
def testProbabilityOfDatasetPlausibility(self):
num_samples = 100
nComps = self.model.GetNumberOfPrincipalComponents()
p_mean = self.model.ComputeProbability(self.model.DrawMean())
for i in xrange(0, num_samples):
coeffs = randn(nComps)
s = self.model.DrawSample(coeffs)
p = self.model.ComputeProbability(s)
# as the distribution we are looking for is just a standard mv normal, all the
# components are independent. We can thus use a 1d normal distribution to compute the
# probability of the full pdf.
pScipy = 1;
for c in coeffs:
pScipy *= normdist(0, 1).pdf(c)
self.assertTrue(p, pScipy)