import sys
sys.path.insert(0,'demos/')
from diffusion import ContaminantTransportModelUpperLeft
# Number of dimensions
num_dim = 3
# The number of components to use for the mixture
num_comp = 1
#-------- The (hypothetical) joint distribution ----------------
# The prior
collection = [UniformND(1), UniformND(1), GammaPDF(1,0.05,1)]
prior = PDFCollection(collection)
# The data
data = np.load('data_concentrations_upperleft_corner.npy')
# The forward model
diff_model = ContaminantTransportModelUpperLeft()
print 'Num_input'
print str(diff_model.num_input) + '\n'
# The isotropic Likelihood
IsotropicL = IsotropicGaussianLikelihood(data[:], diff_model)
# The joint
log_p = Joint(IsotropicL, prior)
print 'Target:'
print str(log_p)
# The approximating distribution
comp = [MultivariateNormal(np.random.gamma(10,1,num_dim)), MultivariateNormal(np.random.gamma(10,1,num_dim))]
from vuq import PDFCollection from vuq import GammaPDF from vuq import UniformND # The number of dimensions num_input = 10 # The parameters # Alpha is chosen uniformly from [0,10] alpha = 10 * np.random.rand(num_input) # Beta is chosen uniformly from [0,1] beta = np.random.rand(num_input) # Make a collection of Gammas: #log_p = PDFCollection([ GammaPDF(alpha[i], beta[i], 1) for i in xrange(num_input) ]) log_p = PDFCollection([UniformND(1) for i in xrange(num_input)]) print str(log_p) print 'Evaluating at many points' print '-' * 80 #x = np.random.gamma(10, 1, [20, num_input]) x = np.random.rand(20, num_input) y = log_p(x) print 'Shape:', y.shape print 'log pdf:' print str(y) + '\n' # Test the evaluation of the gradient at many inputs simultaneously print 'Evaluating gradient at many points' print '-' * 80
from vuq import FirstOrderEntropyApproximation from vuq import ThirdOrderExpectationFunctional from vuq import EvidenceLowerBound from vuq import FullOptimizer from demos import TestModel1 # The number of components to use for the mixture num_comp = 5 # The model model = TestModel1() # The prior log_p_x = MultivariateNormal(mu=[0]) log_p_z_fake = FlatPDF(model.num_output) log_p_x_ext = PDFCollection([log_p_x, log_p_z_fake]) # The isotropic Likelihood log_p_z_given_x = UncertaintyPropagationLikelihood(model, alpha=1e-1) # The joint log_p = Joint(log_p_z_given_x, log_p_x_ext) # The approximating distribution log_q = MixtureOfMultivariateNormals.create(log_p.num_dim, num_comp) # Build the ELBO # Pick an entropy approximation entropy = FirstOrderEntropyApproximation() # Pick an approximation for the expectation of the joint expectation_functional = ThirdOrderExpectationFunctional(log_p) # Build the ELBO elbo = EvidenceLowerBound(entropy, expectation_functional)
from vuq import GammaPDF from vuq import UniformND # The number of dimensions num_input = 10 # The parameters # Alpha is chosen uniformly from [0,10] alpha = 10*np.random.rand(num_input) # Beta is chosen uniformly from [0,1] beta = np.random.rand(num_input) # Make a collection of Gammas: #log_p = PDFCollection([ GammaPDF(alpha[i], beta[i], 1) for i in xrange(num_input) ]) log_p = PDFCollection([ UniformND(1) for i in xrange(num_input) ]) print str(log_p) print 'Evaluating at many points' print '-' * 80 #x = np.random.gamma(10, 1, [20, num_input]) x = np.random.rand(20, num_input) y = log_p(x) print 'Shape:', y.shape print 'log pdf:' print str(y) + '\n' # Test the evaluation of the gradient at many inputs simultaneously print 'Evaluating gradient at many points'