def CDist(scm_a, scm_b, nb_samples_x, nb_samples_k=5, discrete=False):
nodes_a = scm_a.nodes()
nodes_b = scm_b.nodes()
assert nodes_a == nodes_b, \
"The two causal graphs should have the same nodes"
n = len(nodes_a)
x_distrib = IndependentComponentsDistribution(
[UniformDistribution(0, 1) for _ in range(n)])
scm_a.precompute_log_probabilities()
scm_b.precompute_log_probabilities()
results = IDist(scm_a, scm_b, nb_samples_k, discrete=discrete)
for x in x_distrib.sample(nb_samples_x):
# print('x: ' + str(x))
cf_a = en.NUTSSampler(n, scm_a.log_probability)
cf_b = en.NUTSSampler(n, scm_b.log_probability)
counterfactual_scm_a = copy.deepcopy(scm_a)
counterfactual_scm_b = copy.deepcopy(scm_b)
# print("Creating noise")
a_samples = cf_a.sample(x, M=nb_samples_k, Madapt=200)
b_samples = cf_b.sample(x, M=nb_samples_k, Madapt=200)
counterfactual_scm_a.noises_distribution = CounterfactualNoise(
a_samples, x, nb_samples_k)
counterfactual_scm_b.noises_distribution = CounterfactualNoise(
b_samples, x, nb_samples_k)
# print('End noise')
# print("Run IDist")
results += IDist(counterfactual_scm_a,
counterfactual_scm_b,
nb_samples_k,
discrete=discrete)
# print("E nd IDist")
return results / float(nb_samples_x + 1)
def __init__(self, funcs):
"""
A multidimensional freeform probability distribution object for
independent dimensions
Parameters
----------
funcs: list, chippr.gauss or chippr.discrete or chippr.gmix objects
1D functions, one per dimension
"""
self.funcs = funcs
self.dims = len(funcs)
#for d in range(self.dims):
# print('multi dist '+str((d, type(self.funcs[d]))))
self.funcs = [func.dist for func in self.funcs]
self.dist = ICD(self.funcs)
def non_gaussian_noise(n):
IndependentComponentsDistribution([
GammaDistribution(alpha=np.random.rand(), beta=np.random.rand())
for _ in range(n)
])
def default_gaussian_noise(n):
return IndependentComponentsDistribution(
[NormalDistribution(0, np.random.rand()) for _ in range(n)])
def default_noise(n):
return IndependentComponentsDistribution(
[NormalDistribution(0, 0.1) for _ in range(n)])
class multi_dist(object):
def __init__(self, funcs):
"""
A multidimensional freeform probability distribution object for
independent dimensions
Parameters
----------
funcs: list, chippr.gauss or chippr.discrete or chippr.gmix objects
1D functions, one per dimension
"""
self.funcs = funcs
self.dims = len(funcs)
#for d in range(self.dims):
# print('multi dist '+str((d, type(self.funcs[d]))))
self.funcs = [func.dist for func in self.funcs]
self.dist = ICD(self.funcs)
def pdf(self, points):
return self.evaluate(points)
def evaluate_one(self, point):
"""
Function to evaluate the probability at a point in multidimensional
space
Parameters
----------
point: numpy.ndarray, float
coordinate at which to evaluate the probability
Returns
-------
prob: float
probability associated with point
"""
# prob = 1.
# for d in range(self.dims):
# prob *= self.funcs[d].evaluate_one(point[d])
prob = self.dist.probability(point)
return prob
def evaluate(self, points):
"""
Function to evaluate the probability at a point in multidimensional
space
Parameters
----------
points: numpy.ndarray, float
coordinates at which to evaluate the probability
Returns
-------
probs: float
probabilities associated with points
"""
# if len(np.shape(points.T)) == 1:
# return self.evaluate_one(points)
# probs = np.ones(len(points))
# points = points.T
# for d in range(self.dims):
# probs *= self.funcs[d].evaluate(points[d])
probs = self.dist.probability(points)
return probs
def sample_one(self):
"""
Function to take one sample from independent distributions
Returns
-------
point: numpy.ndarray, float
single sample from independent distributions
"""
# point = np.zeros(self.dims)
# for d in range(self.dims):
# point[d] = self.funcs[d].sample_one()
point = self.dist.sample(1)
return point
def sample(self, n_samps):
"""
Function to sample from multivariate probability distribution with independent components
Parameters
----------
n_samps: positive int
number of samples to take
Returns
-------
points: ndarray, float
array of n_samps samples from independent distributions
"""
# print('multi_dist trying to sample '+str(n_samps)+' from '+str(self.dist))
# points = np.array([self.sample_one() for n in range(n_samps)])
points = np.array([self.dist.sample() for n in range(n_samps)])
# print('multi_dist sampled '+str(n_samps)+' from '+str(self.dist))
return points