def enumeration_infer(self, X, e=None) -> ProbDist: #page 525-528
"""
Return the conditional probability distribution of variable X
given evidence e
"""
'''
Use evidence passed to function call, otherwise use default
'''
if e == None:
e = self.evidence
assert X not in e, 'Query variable must be distinct from evidence'
assert X in self.net.variables, 'Variable needs to be in network'
"""
* Initialize a probability distribution
* For each outcome sum over all the hidden variables
* Normalize
"""
Q = ProbDist(X)
# print(Q)
# print('im here')
# for each value xi of X do
for xi in self.net.variable_values(X):
# e[xi] = X
exi = self.copy_assign(e, X, xi)
# print(exi,'exiGHGHGHGHGHHGHGHGHGHGHG')
# Q(xi)←ENUMERATE-ALL(bn.VARS, exi )
Q[xi] = self.enumerate_all(self.net.variables, exi)
# print(e)
"""YOUR CODE"""
# return NORMALIZE(Q(X))
# print(Q, "this is Q")
return Q.normalize()
def rejection_sampling(self,
X: str,
N: int,
e: dict = None) -> ProbDist: #533
"""
Estimate the probability distribution of variable X given using
N samples and evidence e. If not evidence is given, use the default
for the Inference object.
"""
if e == None:
e = self.evidence
# Q = ProbDist(X)
"""YOUR CODE"""
countV = ProbDist(X)
for j in range(N): #N is the sample spaces.
x = self.sample() # x←PRIOR-SAMPLE(bn)
# px=ProbDist(x)
#new way. want to find if e is subset of x items.
setx = set(x.items())
sete = set(e.items())
if (sete.issubset(setx)): #yay
if x[X] is True:
countV[x[X]] = countV[x[X]] + 1
if (x[X] is False):
countV[x[X]] = countV[x[X]] + 1
# print(ProbDist.normalize(countV))
return ProbDist.normalize(countV) #<-- fix this too
def likelihood_weighting(self,
X: str,
N: int,
e: dict = None) -> ProbDist: #page 534
"""Estimate the probability distribution of variable X given
evidence e
"""
if e == None:
e = self.evidence
W = ProbDist(X)
for j in range(N):
x, w = self.weighted_sample(e.copy())
W[x[X]] = W[x[X]] + w
return ProbDist.normalize(W)
def likelihood_weighting(self, X: str, N: int, e: dict = None) -> ProbDist:
"""
TODO: Estimate the probability distribution of variable X given evidence e.
X = query variable
N = total number of samples generated
e = evidence as event
need bayesian net
"""
"""YOUR CODE"""
if e == None:
e = self.evidence
W = {True: 0.0, False: 0.0}
for i in range(N):
x, w = self.weighted_sample(e)
sam = x[X]
W[sam] += w
return ProbDist(X, W) # <- put parameters in it
def rejection_sampling(self, X: str, N: int, e: dict = None) -> ProbDist:
"""
TODO: Estimate the probability distribution of variable X given using
N samples and evidence e. If not evidence is given, use the default
for the Inference object.
X = query variable
N = total number of samples generated
e = evidence as event
need bayesian net
return P(X|e)
"""
if e == None:
e = self.evidence
"""YOUR CODE"""
W = {True: 0.0, False: 0.0}
for i in range(N):
s = self.sample()
if self.consistent(s, e):
sam = s[X]
W[sam] += 1
return ProbDist(X, W) # <- put parameters in it
def enumeration_ask(self, X, e=None) -> ProbDist:
"""
TODO: Return the conditional probability distribution of variable X
given evidence e
"""
# Use evidence passed to function call, otherwise use default
if e == None:
e = self.evidence
assert X not in e, 'Query variable must be distinct from evidence'
assert X in self.net.variables, 'Variable needs to be in network'
"""
* Initialize a probability distribution
* For each outcome sum over all the hidden variables
* Normalize
"""
"""YOUR CODE"""
vars = {}
for xi in self.net.variables:
vars[xi] = self.enumerate_all()
return ProbDist().normalize() # <- may need to tweak a bit
There are two ways to initialize distributions
* Give it dictionary where the keys are the element of the sample space
and the values are counte or un-normalized probablities <<-- Will use this one
* Give a sample_space list/tuple of domain values and a list/tuple that defines
the distibution
'''
''' roll virtual dice '''
counts = {}
for i in range(1, 6 + 1):
counts[i] = 0
for rolls in range(100):
roll = round((random.random() * 6 -
0.5)) + 1 #scale and shift so round will work
counts[roll] += 1
print('Raw counts: ' + str(counts))
pdice = ProbDist('Dice', counts)
print('ProbDist Object: ' + str(pdice))
print('Probability Dsitribution: ' + str(pdice.show_approx()))
'''
BayesNodes and BayesNets
Each node is specified with its parent names and the neccesarey CPT
The CPTs are a little tricky. They are a dictionary where the keys are tuples
of possible values from the sample space of the parent domains. The value
is a probability vector that has the same length as the sample space.
'''
testNode = BayesNode(
'Node', 'val1 val2 val3', 'A/B-Parent C/D-Parent', {
('A', 'C'): (0.1, 0.2, 0.7),
('A', 'D'): (0.3, 0.5, 0.2),
('B', 'C'): (0.0, 0.2, 0.8),