def testProductDistributionSharingMarginals(self):
new_distribution = pd.ProbabilityDistribution(3,2)
new_distribution.distribution = np.array([[0.1,0.15],[0.05,0.20],[.2,.3]])
product_distribution = pd.ProbabilityDistribution(3,2)
product_distribution.distribution = np.array([[ 0.0875 , 0.1625],[ 0.0875, 0.1625] ,[ 0.175 , 0.325 ]])
product_distribution_reconstruct = pd.ProbabilityDistribution(3,2)
product_distribution_reconstruct.distribution = new_distribution.params_of_product_distribution_sharing_marginals()
self.assertEqual(product_distribution, product_distribution_reconstruct)
def get_p_eta(self, eta):
'''
Makes a distribution p_eta with uniform marginals and test statistic eta.
'''
l = self.l
k = self.k
prob_dist = pd.ProbabilityDistribution(k, l)
uniform_dist = pd.ProbabilityDistribution(k, l)
prob_dist_path = pdp.probabilityDistributionPath(uniform_dist)
prob_dist_parameters = prob_dist_path.distribution_at_specified_divergence_from_base_pos_t(
eta)
prob_dist.distribution = prob_dist_parameters
return prob_dist
def construct(self):
firstMarginalsAsMatrix = np.matrix(
[self.feasibleMarginals[0], 1.0 - self.feasibleMarginals[0]])
secondMarginalsAsMatrix = np.matrix(
[self.feasibleMarginals[1], 1.0 - self.feasibleMarginals[1]])
#construct the path through the product distribution with those marginals
productDistributionWithFeasibleMarginals = pd.ProbabilityDistribution(
self.k, self.l)
productDistributionWithFeasibleMarginals.setParametersProductDistAssocWithMarginals(
firstMarginalsAsMatrix, secondMarginalsAsMatrix)
return pdp.probabilityDistributionPath(
productDistributionWithFeasibleMarginals)
def __init__(self, probability_distribution_on_path):
self.base_probability_distribution = pd.ProbabilityDistribution()
self.base_probability_distribution.distribution = probability_distribution_on_path.params_of_product_distribution_sharing_marginals(
)
self.t_max = self.t_max()
self.t_min = self.t_min()
#for situations where it is convenient to mark a special p_eta as the base for calculating KL-divergnece:
self.markedProbabilityDist = None
self.binarySearchOnKLDivergence = bS.binarySearch(
tolerance_exp=10,
maxDepth=500,
increasingFunction=self.KL_divergence_at_t)
self.binarySearchOnMinusKLDivergence = bS.binarySearch(
tolerance_exp=10,
maxDepth=500,
increasingFunction=self.minus_KL_divergence_at_t)
def testRaiseNotImplementedError(self):
k, l = 3, 3
prob_dist = pd.ProbabilityDistribution(k, l)
probDistPath = pdp.probabilityDistributionPath(prob_dist)
probDistPath.tOfMarkedDistribution()
def distributionWrappingParameters(self, parameters):
wrappingDistribution = pd.ProbabilityDistribution()
wrappingDistribution.distribution = parameters
return wrappingDistribution
def distribution_at_t_as_distribution(self, t):
"""return a distribution object version of the distribution_at_t"""
k, l = np.shape(self.base_probability_distribution.distribution)
new_distribution = pd.ProbabilityDistribution(k, l)
new_distribution.distribution = self.distribution_at_t(t)
return new_distribution
def testKLDivergenceOfP_gammaFromDist(self):
new_distribution = pd.ProbabilityDistribution(2,2)
new_distribution.distribution = np.array([[0.25,0.25],[0.25,0.25]])
np.testing.assert_almost_equal(new_distribution.KLDivergenceOfP_gammaFromDist(0.001), 0.000999999947487)
def testEmissionProbabilityFromP_eta_ofProductLikeTypeSizeN(self):
new_distribution = pd.ProbabilityDistribution(3,2)
new_distribution.distribution = np.array([[0.1,0.15],[0.05,0.20],[.2,.3]])
product_distribution = pd.ProbabilityDistribution(3,2)
product_distribution.distribution = np.array([[ 0.0875 , 0.1625],[ 0.0875, 0.1625] ,[ 0.175 , 0.325 ]])
np.testing.assert_almost_equal(new_distribution.emissionProbabilityFromP_eta_ofProductLikeTypeSizeN(product_distribution,100), 4.37171106784e-06)
def testKL_divergenceAsBase(self):
new_distribution2 = pd.ProbabilityDistribution(3,2)
new_distribution2.distribution = np.array([[0.1,0.15],[0.05,0.20],[.2,.3]])
another_distribution_params = np.array([ [.25,0], [0,.25], [.25,.25] ])
self.assertEqual(new_distribution2.KL_divergence_as_base(another_distribution_params),
.25*np.log(.25/.1) + .25*np.log(.25/.20) + .25*np.log(.25/.2) + .25*np.log(.25/.3))
def testProductDistributionFormation(self):
product_distribution = pd.ProbabilityDistribution(3,2)
product_distribution.distribution = np.array([[ 0.0875 , 0.1625],[ 0.0875, 0.1625] ,[ 0.175 , 0.325 ]])
product_distribution_reconstruct = pd.ProbabilityDistribution(3,2)
product_distribution_reconstruct.setParametersProductDistAssocWithMarginals(np.matrix([ 0.25 , 0.25, 0.5 ]), np.matrix([ 0.35, 0.65 ]))
self.assertEqual(product_distribution,product_distribution_reconstruct) #uses __eq__ defined in the probabilityDistribution class
def testFirstAndSecondMarginals(self):
new_distribution = pd.ProbabilityDistribution(3,2)
new_distribution.distribution = np.array([[0.1,0.15],[0.05,0.20],[.2,.3]])
np.testing.assert_almost_equal(new_distribution.first_variable_marginals(),np.array([ 0.25 , 0.25, 0.5 ]))
np.testing.assert_almost_equal(new_distribution.second_variable_marginals(), np.array([ 0.35, 0.65 ]))