def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make sure to
handle the special case where all particles have weight 0 after
reweighting based on observation. If this happens, resample particles
uniformly at random from the set of legal positions
(self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell,
self.getJailPosition()
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeUniformly. The total
weight for a belief distribution can be found by calling totalCount
on a Counter object
util.sample(Counter object) is a helper method to generate a sample from
a belief distribution.
You may also want to use util.manhattanDistance to calculate the
distance between a particle and Pacman's position.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** MODIFIED CODE HERE ***"
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
allPossible = util.Counter()
if noisyDistance == None:
allPossible[self.getJailPosition()] = 1.0
else:
allPossible = util.Counter()
for p in self.particleList:
dist = util.manhattanDistance(p, pacmanPosition)
allPossible[p] += emissionModel[dist]
if allPossible.totalCount() == 0:
self.initializeUniformly(gameState)
return
allPossible.normalize()
self.beliefs = allPossible
for i in range(self.numParticles):
newPos = util.sample(self.beliefs)
self.particleList[i] = newPos
def observe(self, observation, gameState):
"Update beliefs based on the given distance observation."
emissionModel = busters.getObservationDistribution(observation)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
weights = util.Counter()
hasNonZero = False
for p in self.particles:
distance = util.manhattanDistance(p, pacmanPosition)
prob = emissionModel[distance]
weights[p] += prob
if(prob > 0):
hasNonZero = True
if (not hasNonZero):
newParticles = []
for i in range(self.numParticles):
ghostPos = random.choice(self.legalPositions)
newParticles.append(ghostPos)
self.particles = newParticles
return
resampledParticles = []
for i in range(self.numParticles):
resampledParticles.append(util.sample(weights))
self.particles = resampledParticles
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
As in elapseTime, to loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position (2 * i + 1, 1),
where "i" is the 0-based index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of 999 (a noisy distance
of 999 will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles.
"""
pacmanPos = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
jailed = [ noisy == 999 for noisy in noisyDistances ]
partials = [ tuple() ] * self.numParticles
for g in xrange(self.numGhosts):
weighted = util.Counter()
if jailed[g]:
# handle the jailed ghost
jailLocation = (2 * g + 1, 1)
for i in xrange(self.numParticles):
partials[i] += (jailLocation, )
continue
for oldAssign, counts in self.sampledCounts.iteritems():
for assign, oldCount in counts.iteritems():
if oldCount <= 0:
continue
trueDistance = util.manhattanDistance(pacmanPos, assign[g])
delta = abs(trueDistance - noisyDistances[g])
if emissionModels[g][trueDistance] > 0 and delta <= MAX_DIST_DELTA:
# no need to normalize by constant
pTrue = math.exp( -delta )
weighted[assign[g]] = oldCount * emissionModels[g][trueDistance] * pTrue / self.proposals[oldAssign][assign]
totalWeight = weighted.totalCount()
if totalWeight != 0: weighted.normalize()
for i in xrange(self.numParticles):
if totalWeight == 0:
# handle the zero weights case
partials[i] += (random.choice(self.legalPositions), )
else:
partials[i] += (util.sample(weighted), )
self.particles = CounterFromIterable(partials)
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make
sure to handle the special case where all particles have weight
0 after reweighting based on observation. If this happens,
resample particles uniformly at random from the set of legal
positions (self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, **all** particles should be updated so
that the ghost appears in its prison cell, self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeUniformly. The total weight
for a belief distribution can be found by calling totalCount on
a Counter object
util.sample(Counter object) is a helper method to generate a sample from
a belief distribution
You may also want to use util.manhattanDistance to calculate the distance
between a particle and pacman's position.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
newBeliefs = util.Counter()
beliefs = self.getBeliefDistribution()
# Special Case #1
# this will handle the case when pacman captures a ghost, it will update the
# particles to reflect this change
if noisyDistance == None:
newBeliefs[self.getJailPosition()] = 1
# go through all the legal positions and calculate their manhattan distance
# to pacman's position
else:
for p in self.legalPositions:
distance = util.manhattanDistance(p, pacmanPosition)
# if the probability of that distance is greater than zero, than recalculate
# the new beliefs as accounting for that probability
if emissionModel[distance] > 0:
newBeliefs[p] = emissionModel[distance] * beliefs[p]
newBeliefs.normalize()
# Special Case #2
# when all the particles have a weight of 0, we reinitialize
if newBeliefs.totalCount() == 0:
self.initializeUniformly(self.numParticles)
return
self.particles = [util.sample(newBeliefs) for _ in range(self.numParticles)]
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
#first special case
for i in xrange(self.numGhosts):
if noisyDistances[i]==None:
for x, y in enumerate(self.particles):
self.particles[x] = self.getParticleWithGhostInJail(y, i)
#create a weighted particle distribution
weightedParticleDistri = util.Counter()
for particle in self.particles:
product = 1
for i in xrange(self.numGhosts):
if noisyDistances[i]!=None:
trueDistance = util.manhattanDistance(particle[i], pacmanPosition)
product *= emissionModels[i][trueDistance]
weightedParticleDistri[particle] += product
# second special case
if weightedParticleDistri.totalCount()==0:
self.initializeParticles()
#change all particles with all the eaten ghosts' positions changed to their respective jail positions
for i in xrange(self.numGhosts):
if noisyDistances[i]==None:
for x, y in enumerate(self.particles):
self.particles[x] = self.getParticleWithGhostInJail(y, i)
else: # resampling
self.particles = [util.sample(weightedParticleDistri) for particle in self.particles]
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make
sure to handle the special case where all particles have weight
0 after reweighting based on observation. If this happens,
resample particles uniformly at random from the set of legal
positions (self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeUniformly. Remember to
change particles to jail if called for.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
# check if all weights are zero
def zeroWeights(weights):
return all(w == 0 for w in weights.values())
prevBelief = self.getBeliefDistribution()
allPossible = util.Counter()
nextParticles = []
# ghost captured
if noisyDistance is None:
jailPosition = self.getJailPosition()
# put ghost to jail
for i in range(self.numParticles):
nextParticles.append(jailPosition)
self.particles = nextParticles
else:
# update beliefs
for pos in self.legalPositions:
trueDistance = util.manhattanDistance(pos, pacmanPosition)
allPossible[pos] += emissionModel[trueDistance] * prevBelief[pos]
# weights all zero
if zeroWeights(allPossible):
self.initializeUniformly(gameState)
else:
# resample particles
for i in range(self.numParticles):
nextParticles.append(util.sample(allPossible))
self.particles = nextParticles
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. After all particles
are generated randomly, any ghosts that are eaten (have noisyDistance of 0)
must be changed to the jail Position. This will involve changing each
particle if a ghost has been eaten.
** Remember ** We store particles as tuples, but to edit a specific particle,
it must be converted to a list, edited, and then converted back to a tuple. Since
this is a common operation when placing a ghost in the jail for a particle, we have
provided a helper method named self.getParticleWithGhostInJail(particle, ghostIndex)
that performs these three operations for you.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
weighted, particles = util.Counter(), list()
for particle in self.particles:
ghostPositions, weight = list(particle), 1
for ghost in range(self.numGhosts):
manhattan = util.manhattanDistance(ghostPositions[ghost], pacmanPosition)
if noisyDistances[ghost] is None:
particle = self.getParticleWithGhostInJail(particle, ghost)
else:
weight *= emissionModels[ghost][manhattan]
weighted[particle] += weight
weighted.normalize()
if weighted.totalCount() != 0:
for i in range(self.numParticles):
particles.append(util.sampleFromCounter(weighted))
else:
self.initializeParticles()
particles = self.particles
for ghost in range(self.numGhosts):
if noisyDistances[ghost] is None:
for i in range(self.numParticles):
particles[i] = self.getParticleWithGhostInJail(self.particles[i], ghost)
self.particles = particles
def observe(self, observation, gameState):
"""
Updates beliefs based on the distance observation and Pacman's position.
The noisyDistance is the estimated manhattan distance to the ghost you are tracking.
The emissionModel below stores the probability of the noisyDistance for any true
distance you supply. That is, it stores P(noisyDistance | TrueDistance).
self.legalPositions is a list of the possible ghost positions (you
should only consider positions that are in self.legalPositions).
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
# Replace this code with a correct observation update
updatedBeliefs = util.Counter()
if(noisyDistance != 999):
for p in self.legalPositions:
trueDistance = util.manhattanDistance(p, pacmanPosition)
if emissionModel[trueDistance] > 0:
updatedBeliefs[p] = emissionModel[trueDistance] * self.beliefs[p]
else:
# Must take care of this base case - as autograder seems to ask for it.
# Setting the probability of the ghost jail cell to be 1 and rest to be 0
ghostJailPos = (2 * self.ghostAgent.index - 1, 1)
updatedBeliefs[ghostJailPos] = 1.0
# MUST normalize !
updatedBeliefs.normalize()
self.beliefs = updatedBeliefs
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make sure to
handle the special case where all particles have weight 0 after
reweighting based on observation. If this happens, resample particles
uniformly at random from the set of legal positions
(self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell,
self.getJailPosition()
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeUniformly. The total
weight for a belief distribution can be found by calling totalCount
on a Counter object
util.sample(Counter object) is a helper method to generate a sample from
a belief distribution.
You may also want to use util.manhattanDistance to calculate the
distance between a particle and Pacman's position.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
# First check if we ate the ghost, make all particles be the jail position
if noisyDistance is None:
self.particles = [self.getJailPosition() for i in range(self.numParticles)]
else:
new_particles = []
# This will update our beliefs
weights = util.Counter()
# For old particle, use the emission data to inform the new particles
counted = Counter(self.particles)
FoundElt = 0
for key in counted:
distance = util.manhattanDistance(key, pacmanPosition)
# Basically the weight to sample at from at different locations is
# the number of particles at that location * P(noisydist | true dist)
weights[key] = counted[key] * emissionModel[distance]
if weights[key] > 0:
FoundElt = 1
weights.normalize()
# If we didn't get all zeroes
if FoundElt:
# Forward sample, only consider emission
self.particles = []
for i in range(self.numParticles):
self.particles.append(util.sample(weights))
else:
self.initializeUniformly(gameState)
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
indexOfGhosts = [i for i in range(self.numGhosts) if noisyDistances[i] == None]
allPos = util.Counter()
for elem in self.particles:
for i in range (len(indexOfGhosts)):
elem = self.getParticleWithGhostInJail(elem, indexOfGhosts[i])
prob = 1
for i in range(self.numGhosts):
if noisyDistances[i] is not None:
distance = util.manhattanDistance(elem[i], pacmanPosition)
prob *= emissionModels[i][distance]
allPos[elem] += prob
allPos.normalize()
self.beliefs = allPos
if allPos.totalCount() == 0:
self.initializeParticles()
else:
for i in range(len(self.particles)):
self.particles[i] = util.sample(self.beliefs)
def observe(self, observation, gameState):
"""
Updates beliefs based on the distance observation and Pacman's position.
The noisyDistance is the estimated manhattan distance to the ghost you are tracking.
The emissionModel below stores the probability of the noisyDistance for any true
distance you supply. That is, it stores P(noisyDistance | TrueDistance).
aka given the observation noisyDistance, what is the prob of TrueDistance?
self.legalPositions is a list of the possible ghost positions (you
should only consider positions that are in self.legalPositions).
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
# print
# print "TEST"
# print
newBeliefs = util.Counter()
for p in self.beliefs: #for p in self.beliefs
trueDis = util.manhattanDistance(p, pacmanPosition) #get the true distance from pacman to that position
newBeliefs[p] = emissionModel[trueDis]*self.beliefs[p] #inference equation = old belief for that position times current belief for that position in the emission model
newBeliefs.normalize() #normalize that probability
self.beliefs = newBeliefs
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
As in elapseTime, to loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. Remember to
change ghosts' positions to jail if called for.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
#print "pacmanPostion=",pacmanPosition
#print "noisyDistances = ",noisyDistances
#print "emissionModels = ",emissionModels
"*** YOUR CODE HERE ***"
beliefs = self.getBeliefDistribution()
#print "beliefs = ",beliefs
for i in range(self.numGhosts):
if noisyDistances[i] != None:
for particle in beliefs:
ghostItemPos = particle[i]
trueDis = util.manhattanDistance(ghostItemPos,pacmanPosition)
beliefs[particle]*=emissionModels[i][trueDis]
for i in range(self.numGhosts):
if noisyDistances[i] == None:
new_dis = util.Counter()
for key in beliefs:
#print "key=",key
newKey = []
for j in range(self.numGhosts):
if j == i:
newKey.append(self.getJailPosition(i))
else:
newKey.append(key[j])
#print "newKey is ",newKey
new_dis[tuple(newKey)]+=beliefs[key]
beliefs = new_dis
if beliefs.totalCount() == 0:
self.initializeParticles()
else:
for i in range(self.numParticles):
self.particles[i] = util.sample(beliefs)
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
updated = []
for i in range(self.numGhosts):
# print(noisyDistances)
if noisyDistances[i] == None:
for index, particle in enumerate(self.particleList):
self.particleList[index] = self.getParticleWithGhostInJail(particle, i)
weights = util.Counter()
for particle in self.particleList:
totalprob = 1
for i in range(self.numGhosts):
if noisyDistances[i] != None:
dist = util.manhattanDistance(particle[i], pacmanPosition)
prob = emissionModels[i][dist]
totalprob = totalprob * prob
weights[particle] += totalprob
# print(weights)
weights.normalize()
keys = weights.keys()
distribution = weights.values()
if all(val == 0 for val in distribution):
self.initializeParticles()
else:
self.particleList = util.nSample(distribution, keys, len(self.particleList))
def observe(self, observation, gameState):
"Update beliefs based on the given distance observation."
emissionModel = busters.getObservationDistribution(observation)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
noisyDistance = observation
newBeliefs = util.Counter()
beliefs = self.getBeliefDistribution()
for pos in self.legalPositions:
trueDistance = util.manhattanDistance(pacmanPosition, pos)
if emissionModel[trueDistance] > 0:
newBeliefs[pos] = emissionModel[trueDistance] * beliefs[pos]
newBeliefs.normalize()
if newBeliefs.totalCount() == 0:
self.initializeUniformly(gameState, self.numParticles)
else:
self.particles = []
for i in range (0, self.numParticles):
self.particles.append(util.sampleFromCounter(newBeliefs))
if noisyDistance == 999:
self.particles = []
for i in range (0, self.numParticles):
self.particles.append(self.getJailPosition())
def observe(self, observation, gameState):
"""
Updates beliefs based on the distance observation and Pacman's position.
The noisyDistance is the estimated manhattan distance to the ghost you are tracking.
The emissionModel below stores the probability of the noisyDistance for any true
distance you supply. That is, it stores P(noisyDistance | TrueDistance).
self.legalPositions is a list of the possible ghost positions (you
should only consider positions that are in self.legalPositions).
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
# Replace this code with a correct observation update
allPossible = util.Counter()
for p in self.legalPositions:
trueDistance = util.manhattanDistance(p, pacmanPosition)
if emissionModel[trueDistance] > 0:
allPossible[p] = self.beliefs[p]
allPossible[p] *= emissionModel[trueDistance]
allPossible.normalize()
self.beliefs = allPossible
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make
sure to handle the special case where all particles have weight
0 after reweighting based on observation. If this happens,
resample particles uniformly at random from the set of legal
positions (self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, **all** particles should be updated so
that the ghost appears in its prison cell, self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeUniformly. The total weight
for a belief distribution can be found by calling totalCount on
a Counter object
util.sample(Counter object) is a helper method to generate a sample from
a belief distribution
You may also want to use util.manhattanDistance to calculate the distance
between a particle and pacman's position.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
'''
print "Noisy Distance: ", noisyDistance
print "Emission Model: ", emissionModel
print "pacman Position: ", pacmanPosition
print "self.particles: ", self.particles
'''
allPossible = util.Counter()
belief = self.getBeliefDistribution()
#ghost has been captured
if noisyDistance is None:
self.particles = []
for index in range(0, self.numParticles):
self.particles.append(self.getJailPosition())
else:
#weighting and testing if the weights are 0
for position in self.legalPositions:
trueDistance = util.manhattanDistance(position, pacmanPosition)
allPossible[position] = emissionModel[trueDistance] * belief[position]
if allPossible.totalCount() == 0:
self.initializeUniformly(gameState)
return
#resampling
allPossible.normalize()
self.particles = []
for index in range(0, self.numParticles):
newP = util.sample(allPossible)
self.particles.append(newP)
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make
sure to handle the special case where all particles have weight
0 after reweighting based on observation. If this happens,
resample particles uniformly at random from the set of legal
positions (self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, **all** particles should be updated so
that the ghost appears in its prison cell, self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeUniformly. The total weight
for a belief distribution can be found by calling totalCount on
a Counter object
util.sample(Counter object) is a helper method to generate a sample from
a belief distribution
You may also want to use util.manhattanDistance to calculate the distance
between a particle and pacman's position.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
util.raiseNotDefined()
def observe(self, observation, gameState):
"""
Updates beliefs based on the distance observation and Pacman's position.
The noisyDistance is the estimated manhattan distance to the ghost you are tracking.
The emissionModel below stores the probability of the noisyDistance for any true
distance you supply. That is, it stores P(noisyDistance | TrueDistance).
self.legalPositions is a list of the possible ghost positions (you
should only consider positions that are in self.legalPositions).
A correct implementation will handle the following special case:
* When a ghost is captured by Pacman, all beliefs should be updated so
that the ghost appears in its prison cell, position self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
# Replace this code with a correct observation update
# Be sure to handle the jail.
if noisyDistance is None:
self.beliefs = util.Counter()
self.beliefs[self.getJailPosition()]=1.0
else:
for p in self.legalPositions:
trueDistance = util.manhattanDistance(p, pacmanPosition)
self.beliefs[p] = emissionModel[trueDistance]*self.beliefs[p]
self.beliefs.normalize()
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
As in elapseTime, to loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position (2 * i + 1, 1),
where "i" is the 0-based index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of 999 (a noisy distance
of 999 will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make
sure to handle the special case where all particles have weight
0 after reweighting based on observation. If this happens,
resample particles uniformly at random from the set of legal
positions (self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeUniformly. Remember to
change particles to jail if called for.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
util.raiseNotDefined()
def observe(self, observation, gameState):
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
weights = util.Counter()
i = 0
if noisyDistance == None:
self.particle_positions = []
while i != self.numParticles:
self.particle_positions.append(self.getJailPosition())
i += 1
else:
for p in self.particle_positions:
trueDistance = util.manhattanDistance(p, pacmanPosition)
#maybe more than one particle per position
weights[p] += emissionModel[trueDistance]
if weights.totalCount() != 0:
self.particle_positions = []
while i != self.numParticles:
self.particle_positions.append(util.sample(weights))
i += 1
else:
self.initializeUniformly(gameState)
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
counter = util.Counter()
for particle in self.particles:
# weight = 0
weight = 1
for i in range(0, self.numGhosts):
if noisyDistances[i] != None:
manDistance = util.manhattanDistance(particle[i],pacmanPosition)
weight = weight * emissionModels[i][manDistance]
else:
listP = list(particle)
listP[i] = self.getJailPosition(i)
particle = tuple(listP)
counter[particle] = counter[particle] + weight
if any(counter.values()):
particles = []
for i in range(0, self.numParticles):
particles.append(util.sample(counter))
self.particles=particles
else:
self.initializeParticles()
def observe(self, observation, gameState):
"""
Updates beliefs based on the distance observation and Pacman's position.
The noisyDistance is the estimated Manhattan distance to the ghost you
are tracking.
The emissionModel below stores the probability of the noisyDistance for
any true distance you supply. That is, it stores P(noisyDistance |
TrueDistance).
self.legalPositions is a list of the possible ghost positions (you
should only consider positions that are in self.legalPositions).
A correct implementation will handle the following special case:
* When a ghost is captured by Pacman, all beliefs should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
# print 'noisyDistance : ', noisyDistance
# print 'emissionModel : ', emissionModel
# print 'pacmanPosition : ', pacmanPosition
# print ''
"*** Q1 CODE STARTS HERE ***"
# util.raiseNotDefined()
# Replace this code with a correct observation update
# Be sure to handle the "jail" edge case where the ghost is eaten
# and noisyDistance is None
# allPossible = util.Counter()
# for p in self.legalPositions:
# trueDistance = util.manhattanDistance(p, pacmanPosition)
# if emissionModel[trueDistance] > 0:
# allPossible[p] = 1.0
allPossibleParticle = util.Counter()
for p in self.legalPositions:
distance = util.manhattanDistance(p, pacmanPosition)
if emissionModel[distance] > 0:
allPossibleParticle[p] = emissionModel[distance] * self.beliefs[p]
if noisyDistance is None:
jailedGhostBelief = util.Counter()
jailedGhostBelief[(2 * self.index - 1, 1)] = 1
allPossibleParticle = jailedGhostBelief
"*** Q1 CODE ENDS HERE ***"
allPossibleParticle.normalize()
self.beliefs = allPossibleParticle
def observeState(self, gameState):
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
weights = util.Counter()
for i in range(len(self.particle_positions)):
temp = 1
for ghost_idx in range(self.numGhosts):
if noisyDistances[ghost_idx] == None:
new_particle = self.getParticleWithGhostInJail(self.particle_positions[i],ghost_idx)
self.particle_positions[i] = new_particle
else:
trueDistance = util.manhattanDistance(self.particle_positions[i][ghost_idx],pacmanPosition)
model = emissionModels[ghost_idx]
temp *= model[trueDistance]
weights[self.particle_positions[i]] += temp
i = 0
if weights.totalCount() != 0:
self.particle_positions = []
while i != self.numParticles:
self.particle_positions.append(util.sample(weights))
i += 1
else:
self.initializeParticles()
def observe(self, observation, gameState):
"""
Updates beliefs based on the distance observation and Pacman's position.
The noisyDistance is the estimated manhattan distance to the ghost you are tracking.
The emissionModel below stores the probability of the noisyDistance for any true
distance you supply. That is, it stores P(noisyDistance | TrueDistance).
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
allPossible = util.Counter()
for p in self.legalPositions:
trueDistance = util.manhattanDistance(p, pacmanPosition)
if emissionModel[trueDistance] > 0:
#allPossible[p] = 1
allPossible[p] = emissionModel[trueDistance]*self.beliefs[p] #strengthen the belief based on the positive probability vales of the possible locations of ghosts
"*** YOUR CODE HERE ***"
allPossible.normalize()
self.beliefs = allPossible
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. After all particles
are generated randomly, any ghosts that are eaten (have noisyDistance of 0)
must be changed to the jail Position. This will involve changing each
particle if a ghost has been eaten.
** Remember ** We store particles as tuples, but to edit a specific particle,
it must be converted to a list, edited, and then converted back to a tuple. Since
this is a common operation when placing a ghost in the jail for a particle, we have
provided a helper method named self.getParticleWithGhostInJail(particle, ghostIndex)
that performs these three operations for you.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances() # before was a single noisyDistance
if len(noisyDistances) < self.numGhosts: return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
# before was a single one: emissionModel = busters.getObservationDistribution(noisyDistance)
"*** YOUR CODE HERE ***"
newWeightCounter = util.Counter()
for particle in self.particleList:
newWeightForParticle = 1
for index in range(0, self.numGhosts):
if(noisyDistances[index] == None):
particle = self.getParticleWithGhostInJail(particle, index)
else:
ghostsDistanceProduct = util.manhattanDistance(particle[index], pacmanPosition)
newWeightForParticle *= emissionModels[index][ghostsDistanceProduct]
newWeightCounter[particle] = newWeightCounter[particle] + newWeightForParticle
if(not newWeightCounter.totalCount() == 0):
newWeightCounter.normalize()
for x in range(0, self.numParticles):
self.particleList[x] = util.sample(newWeightCounter)
else: #all particles receive 0 weight
self.initializeParticles()
for x in range(0, len(self.particleList)):
for index in range(0, self.numGhosts):
if(noisyDistances[index] == None):
self.particleList[x] = self.getParticleWithGhostInJail(self.particleList[x], index)
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make sure to
handle the special case where all particles have weight 0 after
reweighting based on observation. If this happens, resample particles
uniformly at random from the set of legal positions
(self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell,
self.getJailPosition()
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeUniformly. The total
weight for a belief distribution can be found by calling totalCount
on a Counter object
util.sample(Counter object) is a helper method to generate a sample from
a belief distribution.
You may also want to use util.manhattanDistance to calculate the
distance between a particle and Pacman's position.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
# References: Textbook pp. 598-599
# jail edge case
if noisyDistance == None:
pos = self.getJailPosition()
self.particles = [pos]*self.numParticles
return
# Step 1: calculate weightss
weights = util.Counter()
for pos,prob in self.getBeliefDistribution().items():
trueDistance = util.manhattanDistance(pos, pacmanPosition)
weights[pos] = emissionModel[trueDistance]*prob
weights.normalize()
# Step 2: resample based on the weights
if weights.totalCount() == 0:
self.initializeUniformly(gameState) # edge case
else:
distribution = []
values = []
for pos, prob in weights.items():
values.append(pos)
distribution.append(prob)
self.particles = util.nSample(distribution, values, self.numParticles)
def observe(self, observation, gameState):
"""
Updates beliefs based on the distance observation and Pacman's position.
The noisyDistance is the estimated Manhattan distance to the ghost you
are tracking.
The emissionModel below stores the probability of the noisyDistance for
any true distance you supply. That is, it stores P(noisyDistance |
TrueDistance).
self.legalPositions is a list of the possible ghost positions (you
should only consider positions that are in self.legalPositions).
A correct implementation will handle the following special case:
* When a ghost is captured by Pacman, all beliefs should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
"""
print "############"
print noisyDistance
print emissionModel
print pacmanPosition
print self.beliefs
print "############"
"""
# Replace this code with a correct observation update
# Be sure to handle the "jail" edge case where the ghost is eaten
# and noisyDistance is None
allPossible = util.Counter()
if(noisyDistance != None):
priors = self.beliefs
for p in self.legalPositions:
trueDistance = util.manhattanDistance(p, pacmanPosition)
priorBelief = priors[p]
if emissionModel[trueDistance] > 0:
allPossible[p] = priorBelief * emissionModel[trueDistance]
else:
for p in self.legalPositions:
allPossible[p] = 0
allPossible[self.getJailPosition()] = 1
"*** END YOUR CODE HERE ***"
allPossible.normalize()
self.beliefs = allPossible
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
As in elapseTime, to loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. Remember to
change ghosts' positions to jail if called for.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
allPossible = util.Counter()
for oldParticle in self.particles:
tempDist = 1.0
for i in range(self.numGhosts):
if noisyDistances[i] == None:
tempList = list(oldParticle)
tempList[i] = self.getJailPosition(i)
oldParticle = tuple(tempList)
else:
trueDistance = util.manhattanDistance(oldParticle[i], pacmanPosition)
tempDist *= emissionModels[i][trueDistance]
allPossible[oldParticle] += tempDist
allPossible.normalize()
# check if all particles have zero weight
if list(allPossible) == [0]:
self.initializeParticles()
else:
# resample particles
tempList = []
for x in range(self.numParticles):
tempList.append(util.sample(allPossible));
self.particles = tempList
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. After all particles
are generated randomly, any ghosts that are eaten (have noisyDistance of 0)
must be changed to the jail Position. This will involve changing each
particle if a ghost has been eaten.
** Remember ** We store particles as tuples, but to edit a specific particle,
it must be converted to a list, edited, and then converted back to a tuple. Since
this is a common operation when placing a ghost in the jail for a particle, we have
provided a helper method named self.getParticleWithGhostInJail(particle, ghostIndex)
that performs these three operations for you.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
#check for capture
for i in range(self.numGhosts):
if (noisyDistances[i] == None):
# Ghost is captured
newList = []
for par in self.particleList:
newList.append(self.getParticleWithGhostInJail(par, i))
self.particleList = newList
#weight the particles
weightedCounter = util.Counter()
for par in self.particleList:
prob = 1
for i in range(self.numGhosts):
if (noisyDistances[i] == None): continue
#iterative through ghost
trueDistance = util.manhattanDistance(par[i], pacmanPosition)
prob = prob * emissionModels[i][trueDistance]
weightedCounter[par] += prob
if (weightedCounter.totalCount() == 0):
self.initializeParticles()
return
#resample
newList = []
for i in range(self.numParticles):
newList.append(util.sample(weightedCounter))
self.particleList = newList
return
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
# iterate over all particles and then all ghosts
# if ghost noisyDistances is none then it should go to prison jail
# otherwise we should count probability of this ghost being in specific sell based on ebservation
# and multiply all of them and add to counter
allPossible = util.Counter()
for index in range(0, len(self.particles)):
sample = self.particles[index]
probability = 1
for i in range(self.numGhosts):
if noisyDistances[i] is None:
self.particles[index] = self.getParticleWithGhostInJail(
sample, i)
sample = self.particles[index]
else:
trueDistance = util.manhattanDistance(
sample[i], pacmanPosition)
probability = probability * emissionModels[i][trueDistance]
allPossible[sample] += probability
if (allPossible.totalCount() == 0):
self.initializeParticles()
return
# choose randomly from particles based on their weights
self.particles = []
for i in range(self.numParticles):
self.particles.append(util.sample(allPossible))
def observeState(self, gameState):
"""Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When all particles get weight 0 due to the observation,
a new set of particles need to be generated from the initial
prior distribution by calling initializeParticles.
2) Otherwise after all new particles have been generated by
resampling you must check if any ghosts have been captured
by packman (noisyDistances[i] will be None if ghost i has
ben captured).
For each captured ghost, you need to change the i'th component
of every particle (remember that the particles contain a position
for every ghost---so you need to change the component associated
with the i'th ghost.). In particular, if ghost i has been captured
then the i'th component of every particle must be changed so
the i'th ghost is in its prison cell (position self.getJailPosition(i))
Note that more than one ghost might be captured---you need
to ensure that every particle puts every captured ghost in
its prison cell.
self.getParticleWithGhostInJail is a helper method to help you
edit a specific particle. Since we store particles as tuples,
they must be converted to a list, edited, and then converted
back to a tuple. This is a common operation when placing a
ghost in jail. Note that this function
creates a new particle, that has to replace the old particle in
your list of particles.
HINT1. The weight of every particle is the product of the probabilities
of associated with each ghost's noisyDistance observation
HINT2. When computing the weight of a particle by looking at each
ghost's noisyDistance observation make sure you check
if the ghost has been captured. Captured ghost's are ignored
in the weight computation (the particle's component for
the captured ghost is updated the precise position later---so
this corresponds to multiplying the weight by probability 1
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
jIndex = []
for i in range(self.numGhosts):
if noisyDistances[i] is None:
jIndex.append(i)
counter = util.Counter()
for particle in self.particles:
for j in jIndex:
particle = self.getParticleWithGhostInJail(particle,j)
prob = 1
for i in range(self.numGhosts):
if i not in jIndex:
model = emissionModels[i]
ghostPos = particle[i]
dist = util.manhattanDistance(ghostPos,pacmanPosition)
prob = prob * model[dist]
counter[particle] = counter[particle] + prob
self.beliefs = counter
if counter.totalCount() != 0:
self.beliefs.normalize()
for i in range(len(self.particles)):
newPos = util.sample(self.beliefs)
self.particles[i] = newPos
else:
self.initializeParticles()
"*** END YOUR CODE HERE ***"
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. After all particles
are generated randomly, any ghosts that are eaten (have noisyDistance of 0)
must be changed to the jail Position. This will involve changing each
particle if a ghost has been eaten.
** Remember ** We store particles as tuples, but to edit a specific particle,
it must be converted to a list, edited, and then converted back to a tuple. Since
this is a common operation when placing a ghost in the jail for a particle, we have
provided a helper method named self.getParticleWithGhostInJail(particle, ghostIndex)
that performs these three operations for you.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
# damn this list comprehension is self explainatory
jailedGhosts = [
i for i in range(self.numGhosts) if noisyDistances[i] is None
]
allPossible = util.Counter()
for particle in self.particles:
newParticle = particle
# most of jailed ghosts are captured in previous runs, but there might also be
# a new captured ghost so we need to mark it as jailed
for jailed in jailedGhosts:
newParticle = self.getParticleWithGhostInJail(
newParticle, jailed)
# we start with probability 1 and then multiply for each non-jailed ghost
# with the probability of it being at a certain distance
newProb = 1.0
for ghost in range(self.numGhosts):
if ghost not in jailedGhosts:
# same as in exact inference
trueDistance = util.manhattanDistance(
pacmanPosition, newParticle[ghost])
newProb *= emissionModels[ghost][trueDistance]
allPossible[newParticle] += newProb
self.beliefs = allPossible
self.beliefs.normalize()
# make new particles from obtained beliefs
if self.beliefs.totalCount() == 0:
self.initializeParticles()
else:
self.particles = [
util.sample(self.beliefs) for i in range(self.numParticles)
]
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
# Checking if noisyDistance is None i.e prison cell and hence making the probability of the belief of prison cell 1
for i in range(self.numGhosts):
if noisyDistances[i] == None:
self.particles = [
self.getParticleWithGhostInJail(p, i)
for p in self.particles
]
weightedParticles = util.Counter()
allPossible = util.Counter()
#changing the probability of beliefs using formula (new_prob = old_prob * conditional_prob) same as bayes net P(h/D) = P(D/h)*P(h)
for p in self.particles:
weightedParticles[p] = 1
for i in range(self.numGhosts):
if noisyDistances[i] != None:
trueDistance = util.manhattanDistance(p[i], pacmanPosition)
weightedParticles[p] *= emissionModels[i][trueDistance]
weightedParticles.normalize()
self.particleWeights = util.Counter()
for p in self.particles:
self.particleWeights[p] += 1.0
for p in self.particles:
if p not in allPossible:
allPossible[p] = weightedParticles[p] * self.particleWeights[p]
allPossible.normalize()
self.beliefs = allPossible
if self.beliefs.totalCount() == 0:
self.initializeParticles()
for i in range(self.numGhosts):
if noisyDistances[i] == None:
self.particles = [
self.getParticleWithGhostInJail(p, i)
for p in self.particles
]
else:
self.particles = [
util.sample(self.beliefs) for i in range(self.numParticles)
]
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
weight = util.Counter()
for p in self.particles:
particleList = list(p)
particleWeight = 1
for i in range(self.numGhosts):
trueDistance = util.manhattanDistance(particleList[i],
pacmanPosition)
if noisyDistances[i] == None:
p = self.getParticleWithGhostInJail(p, i)
emission = 1
particleWeight *= emission
else:
emission = emissionModels[i]
particleWeight *= emission[trueDistance]
weight[p] += particleWeight
weight.normalize()
tempList = []
tempList2 = []
if weight.totalCount() == 0:
self.initializeParticles()
tempList = self.particles
for i in range(self.numGhosts):
if noisyDistances[i] == None:
for p in range(self.numParticles):
tempList[p] = self.getParticleWithGhostInJail(
self.particles[p], i)
self.particles = tempList
else:
for i in range(self.numParticles):
tempList2.append(util.sampleFromCounter(weight))
self.particles = tempList2
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
currentDistribution = self.getBeliefDistribution()
# Create new distribution model
newDistribution = util.Counter()
for state, stateProb in currentDistribution.iteritems():
newProb = stateProb
for ghostIndex in range(self.numGhosts):
# If ghost is in jail
if noisyDistances[ghostIndex] == None:
state = self.getParticleWithGhostInJail(state, ghostIndex)
else:
trueDistance = util.manhattanDistance(state[ghostIndex], pacmanPosition)
newProb *= emissionModels[ghostIndex][trueDistance]
newDistribution[state] = newProb
# Handle some edge cases
if newDistribution.totalCount() == 0.0:
self.initializeParticles()
newDistribution = self.getBeliefDistribution()
# Using new distribution sample particles
items = sorted(newDistribution.items())
distribution = [i[1] for i in items]
values = [i[0] for i in items]
newListOfSamples = util.nSample(distribution, values, self.numParticles)
self.particles = newListOfSamples
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
def jailGhosts(particle):
newParticle = list(particle)
for ghost in range(self.numGhosts):
if noisyDistances[ghost] is None:
newParticle[ghost] = self.getJailPosition(ghost)
return tuple(newParticle)
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
# Distribution of particles
distribution = util.Counter()
for particle in self.particles:
prob = 1.0
for ghost in range(self.numGhosts):
if noisyDistances[ghost] is not None:
distance = util.manhattanDistance(pacmanPosition,
particle[ghost])
prob *= emissionModels[ghost][distance]
distribution[particle] += prob
# If all particles have 0 weight, recreate from the prior distribution
if distribution.totalCount() == 0:
self.initializeParticles()
self.particles = map(jailGhosts, self.particles)
return
newParticles = []
for i in range(self.numParticles):
particle = util.sample(distribution)
particle = jailGhosts(particle)
newParticles.append(tuple(particle))
self.particles = newParticles
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. After all particles
are generated randomly, any ghosts that are eaten (have noisyDistance of None)
must be changed to the jail Position. This will involve changing each
particle if a ghost has been eaten.
** Remember ** We store particles as tuples, but to edit a specific particle,
it must be converted to a list, edited, and then converted back to a tuple. Since
this is a common operation when placing a ghost in the jail for a particle, we have
provided a helper method named self.getParticleWithGhostInJail(particle, ghostIndex)
that performs these three operations for you.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
# instanciate grid world
grid = util.Counter()
for particle in self.particles:
prob = 1.0
# loop throgh ghots
for index in range(self.numGhosts):
# ghost is in jail
if noisyDistances[index] == None:
particle = self.getParticleWithGhostInJail(particle, index)
# get distance from ghost and update prob
elif prob != 0:
man_distance = util.manhattanDistance(particle[index], pacmanPosition)
prob = prob * emissionModels[index][man_distance]
# update grid
grid[particle] = grid[particle] + prob
# all ghosts are in jail
if sum(grid.values()) == 0:
self.initializeParticles()
return
# normalyze geid world and instanciate relevant
grid.normalize()
self.particles = []
counter = 0
# resample particles based on state of the grid
while counter < self.numParticles:
particle = util.sample(grid)
self.particles.append(particle)
counter = counter + 1
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
As in elapseTime, to loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. Remember to
change ghosts' positions to jail if called for.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
"""allPossible = util.Counter()
for p in self.particlesList:
for i in range(self.numGhosts):
if noisyDistances[i] == None:
listP = list(particle)
listP[g] = self.getJailPosition(g)
particle = tuple(listP)
else:
trueDistance = util.manhattanDistance(p[i], pacmanPosition)
weight *= emissionModels[i][trueDistance]
allPossible[p] += weight
if allPossible.totalCount() == 0:
self.initializeUniformly(gameState)
else:
self.particlesList = []
for i in range(self.numParticles):
self.particlesList.append(util.sample(allPossible))
"""
beliefDis = self.getBeliefDistribution()
for i in range(self.numGhosts):
if noisyDistances[i] != None:
for particle in beliefDis:
ghost_pos = particle[i]
trueDistance = util.manhattanDistance(
ghost_pos, pacmanPosition)
beliefDis[particle] *= emissionModels[i][trueDistance]
for i in range(self.numGhosts):
if noisyDistances[i] == None:
new_dis = util.Counter()
for particle in beliefDis:
new_dis[self.getParticleWithGhostInJail(
particle, i)] += beliefDis[particle]
beliefDis = new_dis
if beliefDis.totalCount() == 0:
self.initializeParticles()
else:
for i in xrange(self.numParticles):
self.particles[i] = util.sample(beliefDis)
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
allPossible = util.Counter()
#updates the weight for each particle
for part in self.particles:
weight = 1
for i in range(self.numGhosts):
#ghost in jail, changes particle location to jail
if noisyDistances[i] == None:
part = self.getParticleWithGhostInJail(part, i)
else:
#ghost not in jail
#updates weightrs
weight = weight * emissionModels[i][util.manhattanDistance(
part[i], pacmanPosition)]
#if ghost in jail, weight is now in Jail
#if ghost is not in jail, weight is updated nomrally with the particle
allPossible[part] = allPossible[part] + weight
#checks if all particles are valued at 0, and resets
if sum(allPossible.values()) == 0:
self.initializeParticles()
else:
allPossible.normalize()
self.particles = []
for i in range(self.numParticles):
self.particles.append(tuple(util.sample(allPossible)))
def observeState(self, gameState):
"""Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When all particles get weight 0 due to the observation,
a new set of particles need to be generated from the initial
prior distribution by calling initializeParticles.
2) Otherwise after all new particles have been generated by
resampling you must check if any ghosts have been captured
by packman (noisyDistances[i] will be None if ghost i has
ben captured).
For each captured ghost, you need to change the i'th component
of every particle (remember that the particles contain a position
for every ghost---so you need to change the component associated
with the i'th ghost.). In particular, if ghost i has been captured
then the i'th component of every particle must be changed so
the i'th ghost is in its prison cell (position self.getJailPosition(i))
Note that more than one ghost might be captured---you need
to ensure that every particle puts every captured ghost in
its prison cell.
self.getParticleWithGhostInJail is a helper method to help you
edit a specific particle. Since we store particles as tuples,
they must be converted to a list, edited, and then converted
back to a tuple. This is a common operation when placing a
ghost in jail. Note that this function
creates a new particle, that has to replace the old particle in
your list of particles.
HINT1. The weight of every particle is the product of the probabilities
of associated with each ghost's noisyDistance observation
HINT2. When computing the weight of a particle by looking at each
ghost's noisyDistance observation make sure you check
if the ghost has been captured. Captured ghost's are ignored
in the weight computation (the particle's component for
the captured ghost is updated the precise position later---so
this corresponds to multiplying the weight by probability 1
"""
# emissionModel[dist(p)] = Pr(et|xt=p).
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
# for each distinct particle, compute weight as sum of weights of all the same particle
# for each particle, compute weight as product of porobability of each postion in that particle.
weights = util.Counter()
for particle in self.particles:
product = 1.0
for i in range(self.numGhosts):
if noisyDistances[i] == None:
particle = self.getParticleWithGhostInJail(particle, i)
else:
product *= emissionModels[i][util.manhattanDistance(
particle[i], pacmanPosition)]
weights[particle] += product
if weights.totalCount() > 0:
weights.normalize()
for i in range(self.numParticles):
self.particles[i] = util.sample(weights)
else: # case 1
self.initializeParticles()
for index, particle in enumerate(self.particles): # case 2
for i in range(self.numGhosts):
if noisyDistances[i] == None:
self.particles[index] = self.getParticleWithGhostInJail(
particle, i)
"*** END YOUR CODE HERE ***"
def observe(self, observation, gameState):
"""Updates beliefs based on the distance observation and Pacman's
position.
When we enter this function pacman's distribution over
possible locations of the ghost are stored in self.beliefs
For any position p:
self.beliefs[p] = Pr(Xt=p | e_{t-1}, e_{t-2}, ..., e_1)
That is, pacman's distribution has already been updated by all
prior observations already.
This function should update self.beliefs[p] so that
self.beliefs[p] = Pr(Xt=p |e_t, e_{t-1}, e_{t-2}, ..., e_1)
That is, it should update pacman's distribution over the
ghost's locations to account for the passed observation.
noisyDistance (= the next observation e_t) is the estimated
Manhattan distance to the ghost you are tracking.
emissionModel = busters.getObservationDistribution(noisyDistance)
stores the probability of having observed noisyDistance given any
true distance you supply. That is
emissionModel[trueDistance] = Pr(noisyDistance | trueDistance).
Since our observations have to do with manhattanDistance with
no indication of direction, we take
Pr(noisyDistance | Xt=p) =
Pr(noisyDistance | manhattanDistance(p,packmanPosition))
That is, the probability of observing noisyDistance given that the
ghost is in position p is equal to the probability of having
observed noisyDistance given the trueDistance between p and the
pacman's current position.
self.legalPositions is a list of the possible ghost positions
(Only positions in self.legalPositions need to have
their probability updated)
A correct implementation will handle the following special
case:
* When a ghost is captured by Pacman, all beliefs should be
updated so that pacman believes the ghost to be in its
prison cell with probability 1, this position is
self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured, note 0 != None).
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
#the code below updates pacman's beliefs so that it
#has a uniform distribution over all possible positions
#the ghost could be.
#
# Replace this code with a correct observation update
# Be sure to handle the "jail" edge case where the ghost is eaten
# and noisyDistance is None
#print(self.getJailPosition())
#print(noisyDistance)
#print(emissionModel)
#print(pacmanPosition)
#print(self.beliefs)
#print(self.legalPositions)
# The counter class is an extension of the standard pythoy dictionary type
beliefs = util.Counter()
if noisyDistance == None:
beliefs[self.getJailPosition()] = 1.0
else:
for ghostPosition in self.legalPositions:
distance = util.manhattanDistance(ghostPosition,
pacmanPosition)
# emissionModel[trueDistance] = Pr(noisyDistance | trueDistance) = Pr(et|xt)
# self.beliefs[p] = Pr(Xt=p |e_t, e_{t-1}, e_{t-2}, ..., e_1)
beliefs[ghostPosition] = emissionModel[
distance] * self.beliefs[ghostPosition]
"*** END YOUR CODE HERE ***"
beliefs.normalize()
self.beliefs = beliefs
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
newBeliefDist = util.Counter()
eatenGhostsIndex = []
for i in range(self.numGhosts):
if noisyDistances[i] is None:
eatenGhostsIndex.append(i)
for particle in self.particles:
# Update for eaten Ghosts
for i in eatenGhostsIndex:
particle = self.getParticleWithGhostInJail(particle, i)
prob = 1
# For not eaten Ghosts
for i in range(self.numGhosts):
if i not in eatenGhostsIndex:
trueDistance = util.manhattanDistance(
particle[i], pacmanPosition)
prob *= (emissionModels[i])[trueDistance]
newBeliefDist[particle] += prob
if newBeliefDist.totalCount() == 0:
self.initializeParticles()
else:
newBeliefDist.normalize()
for pos in range(len(self.particles)):
self.particles[pos] = util.sample(newBeliefDist)
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
#first special case
for i in xrange(self.numGhosts):
if noisyDistances[i] == None:
for x, y in enumerate(self.particles):
self.particles[x] = self.getParticleWithGhostInJail(y, i)
#create a weighted particle distribution
weightedParticleDistri = util.Counter()
for particle in self.particles:
product = 1
for i in xrange(self.numGhosts):
if noisyDistances[i] != None:
trueDistance = util.manhattanDistance(
particle[i], pacmanPosition)
product *= emissionModels[i][trueDistance]
weightedParticleDistri[particle] += product
# second special case
if weightedParticleDistri.totalCount() == 0:
self.initializeParticles()
#change all particles with all the eaten ghosts' positions changed to their respective jail positions
for i in xrange(self.numGhosts):
if noisyDistances[i] == None:
for x, y in enumerate(self.particles):
self.particles[x] = self.getParticleWithGhostInJail(
y, i)
else: # resampling
self.particles = [
util.sample(weightedParticleDistri)
for particle in self.particles
]
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
# jail edge case
for i in range(self.numGhosts):
if noisyDistances[i] == None:
pos = self.getJailPosition(i)
for j,particle in enumerate(self.particles):
self.particles[j] = self.getParticleWithGhostInJail(particle,i)
# Step 1: calculate weights
weights = util.Counter()
for particle,N in self.getBeliefDistribution().items():
trueDistance = util.manhattanDistance(particle[i], pacmanPosition)
prob = 1.0
for i in range(self.numGhosts):
if noisyDistances[i] != None:
distance = util.manhattanDistance(particle[i], pacmanPosition)
prob *= emissionModels[i][distance]
weights[particle] += prob*N
weights.normalize()
# Step 2: resample based on the weights
if weights.totalCount() == 0:
self.initializeParticles() # edge case
else:
distribution = []
values = []
for pos, prob in weights.items():
values.append(pos)
distribution.append(prob)
self.particles = util.nSample(distribution, values, self.numParticles)
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
allPossible = util.Counter()
for i in range(self.numParticles):
prob = 1
each = self.particleslist[i]
# print("!!!!!!!!!!")
# print(each)
for j in range(self.numGhosts):
noisydis = noisyDistances[j]
emissionmod = emissionModels[j]
if noisydis ==None:
each = self.getParticleWithGhostInJail(each,j)
else:
trueDistance = util.manhattanDistance(each[j], pacmanPosition)
prob *=emissionmod[trueDistance]
allPossible[each]+=prob
allPossible.normalize()
if allPossible.totalCount()==0:
self.initializeParticles()
else:
samplelist = []
count = 0
while count < self.numParticles:
samplelist.append(util.sample(allPossible))
count +=1
self.particleslist = samplelist
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
ghostIndexJail = []
for i in range(self.numGhosts):
if noisyDistances[i] == None:
ghostIndexJail.append(i)
newParticles = util.Counter()
# Creating all particles
for p in self.particles:
for ghosts in ghostIndexJail:
p = self.getParticleWithGhostInJail(p, ghosts)
probability = 1
for i in range(self.numGhosts):
if i not in ghostIndexJail:
em = emissionModels[i]
#true distance
dist = util.manhattanDistance(p[i], pacmanPosition)
probability *= em[dist]
# particle
newParticles[p] += probability
self.beliefs = newParticles
# same as previous question
# If all values are zero .i.e sum should be zero
# then initialze partciles
if newParticles.totalCount() == 0:
self.initializeParticles()
else:
self.beliefs.normalize()
for i in range(len(self.particles)):
newPos = util.sample(self.beliefs)
self.particles[i] = newPos
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make sure to
handle the special case where all particles have weight 0 after
reweighting based on observation. If this happens, resample particles
uniformly at random from the set of legal positions
(self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell,
self.getJailPosition()
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeUniformly. The total
weight for a belief distribution can be found by calling totalCount
on a Counter object
util.sample(Counter object) is a helper method to generate a sample from
a belief distribution.
You may also want to use util.manhattanDistance to calculate the
distance between a particle and Pacman's position.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
allPossible = util.Counter()
weightedParticles = util.Counter()
particlesDistribution = util.Counter()
# Checking if noisyDistance is None i.e prison cell and hence making the probability of the belief of prison cell 1
if noisyDistance == None:
self.particles = [
self.getJailPosition() for i in range(self.numParticles)
]
allPossible[self.getJailPosition()] = 1
self.beliefs = allPossible
else:
#changing the probability of beliefs using formula (new_prob = old_prob * conditional_prob) same as bayes net P(h/D) = P(D/h)*P(h)
for p in self.particles:
trueDistance = util.manhattanDistance(p, pacmanPosition)
allPossible[
p] = self.particleWeights[p] * emissionModel[trueDistance]
weightedParticles[p] += 1
allPossible.normalize()
self.beliefs = allPossible
self.particleWeights = weightedParticles
if allPossible.totalCount() == 0:
self.initializeUniformly(gameState)
else:
self.particles = [
util.sample(self.beliefs) for i in range(self.numParticles)
]
self.particleWeights = util.Counter()
for p in self.particles:
self.particleWeights[p] += 1.0
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
import functools
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
particleWeights = []
#if ghost is eaten place it in jail with a probability of 1
#for p in self.particles:
# for i in range(self.numGhosts):
# if noisyDistances[i] == None:
# p = self.getParticleWithGhostInJail(p, i)
# jailGhost, res = float('inf'), self.doJailCheck(noisyDistances)
# if res[1] is True: jailGhost = res[0]
jailCondition = False
#weighting the particles
for index, p in enumerate(self.particles):
listOfLocationWeights = []
# loop through all ghosts
for i in range(self.numGhosts):
if noisyDistances[i] != None:
# find the true distance from pacman to the current ghost that we're iterating through
trueDistance = util.manhattanDistance(p[i], pacmanPosition)
# weight each particle by the probability of getting to that position (use emission model)
# account for our current belief distribution (evidence) at this point in time
listOfLocationWeights.append(
emissionModels[i][trueDistance]) #* currentBeliefs[p])
else:
self.particles[index] = self.getParticleWithGhostInJail(
p, i)
jailCondition = True
break
#self.particles = [self.getParticleWithGhostInJail(p, i) for p in self.particles]
#continue
if len(listOfLocationWeights) != 0:
particleWeights.append(
functools.reduce(lambda x, y: x * y,
listOfLocationWeights))
else:
particleWeights.append(0)
#if not jailCondition:
# now create a counter and count up the particle weights observed
particleDictionary = util.Counter()
#for i in range(self.numParticles):
for index, p in enumerate(self.particles):
#particleDictionary[self.particles[i]] += particleWeights[i]
particleDictionary[p] += particleWeights[index]
particleDictionary.normalize()
# 2) if all particles have 0 weight, recreate prior distribution
#print particleDictionary.totalCount() == 0
if particleDictionary.totalCount() == 0:
self.initializeParticles()
self.doJailCheck(noisyDistances)
# otherwise, go ahead and resample based on our new beliefs
else:
keys = []
values = []
# find each key, value pair in our counter
keys, values = zip(*particleDictionary.items())
# resample self.particles
self.particles = util.nSample(values, keys, self.numParticles)
self.doJailCheck(noisyDistances)
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make
sure to handle the special case where all particles have weight
0 after reweighting based on observation. If this happens,
resample particles uniformly at random from the set of legal
positions (self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, **all** particles should be updated so
that the ghost appears in its prison cell, self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeUniformly. The total weight
for a belief distribution can be found by calling totalCount on
a Counter object
util.sample(Counter object) is a helper method to generate a sample from
a belief distribution
You may also want to use util.manhattanDistance to calculate the distance
between a particle and pacman's position.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
allPossible = util.Counter()
if noisyDistance is None:
# if ghost is captured, mark it so
allPossible[self.getJailPosition()] = 1.0
else:
for p in self.particles:
# for each particle get the position and add the probability
trueDistance = util.manhattanDistance(pacmanPosition, p)
if emissionModel[trueDistance] > 0:
# add the probability to be at distance trueDistance
allPossible[p] += emissionModel[trueDistance]
# make sure we have sum = 1 over probabilities
allPossible.normalize()
self.beliefs = allPossible
newParticles = []
if self.beliefs.totalCount() == 0:
# this conditional branch is pretty much stated in the requirement
self.initializeUniformly(gameState)
else:
# the next particles will be samples from current beliefs
# if a belief from the current state has a high probability, then it will happen more
# often, and will have more particles in that position
# the more a position appears in the particles, the higher the probability
# that a ghost is there
self.particles = [
util.sample(allPossible) for i in range(len(self.particles))
]
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
As in elapseTime, to loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. Remember to
change ghosts' positions to jail if called for.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
for i in range(self.numGhosts):
if noisyDistances[i] is None:
for ptc in self.particles[:]:
ptc = list(ptc)
ptc[i] = self.getJailPosition(i)
self.particles.append(tuple(ptc))
weighted = util.Counter()
for particle in self.particles:
weighted_ptc = 1.0
for i in range(self.numGhosts):
if noisyDistances[i] is not None:
weighted_ptc *= emissionModels[i][util.manhattanDistance(
pacmanPosition, particle[i])]
weighted[particle] += weighted_ptc
weighted.normalize()
if weighted.totalCount() == 0:
self.initializeParticles()
for i in range(self.numGhosts):
if noisyDistances[i] is None:
for ptc in self.particles[:]:
ptc = list(ptc)
ptc[i] = self.getJailPosition(i)
self.particles.append(tuple(ptc))
else:
self.particles = [
util.sample(weighted) for i in range(self.numParticles)
]
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
As in elapseTime, to loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. After all particles
are generated randomly, any ghosts that are eaten (have noisyDistance of 0)
must be changed to the jail Position. This will involve changing each
particle if a ghost has been eaten.
** Remember ** We store particles as tuples, but to edit a specific particle,
it must be converted to a list, edited, and then converted back to a tuple. Since
this is a common operation when placing a ghost in the jail for a particle, we have
provided a helper method named self.getParticleWithGhostInJail(particle, ghostIndex)
that performs these three operations for you.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
for i in range(self.numGhosts):
if noisyDistances[i] is None:
self.particles = [
self.getParticleWithGhostInJail(particle, i)
for particle in self.particles[:]
]
weightedParticles = util.Counter()
for particle in self.particles:
weightedParticle = 1.0
for i in range(self.numGhosts):
if noisyDistances[i] is not None:
weightedParticle *= emissionModels[i][
util.manhattanDistance(pacmanPosition, particle[i])]
weightedParticles[particle] += weightedParticle
weightedParticles.normalize()
if weightedParticles.totalCount() == 0:
self.initializeParticles()
for i in range(self.numGhosts):
if noisyDistances[i] is None:
self.particles = [
self.getParticleWithGhostInJail(particle, i)
for particle in self.particles[:]
]
else:
self.particles = [
util.sample(weightedParticles)
for i in range(self.numParticles)
]
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
allPossible = util.Counter()
for particle in self.particles:
accrueProb = 1.0
for ghost in range(self.numGhosts):
# handle a ghost in jail by updating the particle
if noisyDistances[ghost] == None:
particle = self.getParticleWithGhostInJail(particle, ghost)
else:
trueDistance = util.manhattanDistance(
particle[ghost], pacmanPosition)
accrueProb *= emissionModels[ghost][trueDistance]
allPossible[particle] += accrueProb
# handle all particles getting a weight of zero by starting again
if allPossible.totalCount() == 0:
self.initializeParticles()
# add back the information for ghosts in jail
for ghost in range(self.numGhosts):
if noisyDistances[ghost] == None:
for j, particle in enumerate(self.particles):
self.particle[j] = self.getParticleWithGhostInJail(
particle, ghost)
else:
# rebuild the particle list for next time!
self.particles = []
allPossible.normalize()
for index in range(self.numParticles):
self.particles.append(util.sample(allPossible))
def observe(self, observation, gameState):
"""Updates beliefs based on the distance observation and Pacman's
position.
When we enter this function pacman's distribution over
possible locations of the ghost are stored in self.beliefs
For any position p:
self.beliefs[p] = Pr(Xt=p | e_{t-1}, e_{t-2}, ..., e_1)
That is, pacman's distribution has already been updated by all
prior observations already.
This function should update self.beliefs[p] so that
self.beliefs[p] = Pr(Xt=p |e_t, e_{t-1}, e_{t-2}, ..., e_1)
That is, it should update pacman's distribution over the
ghost's locations to account for the passed observation.
noisyDistance (= the next observation e_t) is the estimated
Manhattan distance to the ghost you are tracking.
emissionModel = busters.getObservationDistribution(noisyDistance)
stores the probability of having observed noisyDistance given any
true distance you supply. That is
emissionModel[trueDistance] = Pr(noisyDistance | trueDistance).
Since our observations have to do with manhattanDistance with
no indication of direction, we take
Pr(noisyDistance | Xt=p) =
Pr(noisyDistance | manhattanDistance(p,packmanPosition))
That is, the probability of observing noisyDistance given that the
ghost is in position p is equal to the probability of having
observed noisyDistance given the trueDistance between p and the
pacman's current position.
self.legalPositions is a list of the possible ghost positions
(Only positions in self.legalPositions need to have
their probability updated)
A correct implementation will handle the following special
case:
* When a ghost is captured by Pacman, all beliefs should be
updated so that pacman believes the ghost to be in its
prison cell with probability 1, this position is
self.getJailPosition()
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured, note 0 != None).
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
allPossible = util.Counter()
if noisyDistance is None:
allPossible[self.getJailPosition] = 1
else:
for p in self.legalPositions:
allPossible[p] = self.beliefs[p] * \
emissionModel[util.manhattanDistance(
p, pacmanPosition)]
allPossible.normalize()
self.beliefs = allPossible
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [
busters.getObservationDistribution(dist) for dist in noisyDistances
]
"*** YOUR CODE HERE ***"
allPossible = util.Counter()
old_beliefs = self.getBeliefDistribution()
for ind, particle in enumerate(self.particles):
par = 1.0
for i in range(self.numGhosts):
if (noisyDistances[i] == None):
particle = self.getParticleWithGhostInJail(particle, i)
else:
dist = util.manhattanDistance(particle[i], pacmanPosition)
par = par * emissionModels[i][dist]
allPossible[particle] += par
if not any(allPossible.values()):
self.initializeParticles()
for ind, particle in enumerate(self.particles):
for i in range(self.numGhosts):
if (noisyDistances[i] == None):
particle = self.getParticleWithGhostInJail(particle, i)
else:
allPossible.normalize()
temp = []
for i in range(self.numParticles):
temp.append(util.sample(allPossible))
self.particles = temp
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated so
that the ghost appears in its prison cell, position self.getJailPosition(i)
where "i" is the index of the ghost.
You can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None (a noisy distance
of None will be returned if, and only if, the ghost is
captured).
2) When all particles receive 0 weight, they should be recreated from the
prior distribution by calling initializeParticles. After all particles
are generated randomly, any ghosts that are eaten (have noisyDistance of None)
must be changed to the jail Position. This will involve changing each
particle if a ghost has been eaten.
** Remember ** We store particles as tuples, but to edit a specific particle,
it must be converted to a list, edited, and then converted back to a tuple. Since
this is a common operation when placing a ghost in the jail for a particle, we have
provided a helper method named self.getParticleWithGhostInJail(particle, ghostIndex)
that performs these three operations for you.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts: return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
allPossible = util.Counter()
place = 0
for number in self.particles:
remainder = 1.0
for num in range(self.numGhosts):
# examining case
if noisyDistances[num] is None:
number = self.getParticleWithGhostInJail(number, num)
else:
gameRange = util.manhattanDistance(number[num], pacmanPosition)
remainder = remainder * emissionModels[num][gameRange]
allPossible[number] = allPossible[number] + remainder
place = place + 1
if not any(allPossible.values()):
self.initializeParticles()
place = 0
for number in self.particles:
for num in range(self.numGhosts):
# examining case
if noisyDistances[num] is not None:
break
else:
number = self.getParticleWithGhostInJail(number, num)
place = place + 1
else:
allPossible.normalize()
store = []
for number in range(self.numParticles):
store.append(util.sample(allPossible))
self.particles = store
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make sure to
handle the special case where all particles have weight 0 after
reweighting based on observation. If this happens, resample particles
uniformly at random from the set of legal positions
(self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell,
self.getJailPosition()
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeUniformly. The total
weight for a belief distribution can be found by calling totalCount
on a Counter object
util.sample(Counter object) is a helper method to generate a sample from
a belief distribution.
You may also want to use util.manhattanDistance to calculate the
distance between a particle and Pacman's position.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
"*** YOUR CODE HERE ***"
# initialize the counter and retrieve belief distribution
allPossible = util.Counter()
beliefs = self.getBeliefDistribution()
# Case 1
# update all particles when ghost is eaten (noisyDistance == None)
if noisyDistance == None:
# do this as a list because our particles are in a list
jailPositionList = [self.getJailPosition()]
self.particlesList = self.numParticles * jailPositionList
else:
# similar to q2
for p in self.legalPositions:
# retrieve true distance to each location and noisy
trueDistance = util.manhattanDistance(p, pacmanPosition)
probabilityOfNoisyGivenTrue = emissionModel[trueDistance]
# we want to keep summing them according to the particle
allPossible[p] = allPossible[p] + (beliefs[p] * probabilityOfNoisyGivenTrue)
# Check whether particles all have a weight of 0 and initialize uniformly if true
if allPossible.totalCount() == 0:
self.initializeUniformly(gameState)
else:
# iterate through particles
particleSampleDistribution = list()
# create the samples from the distribution based on new counter
for _ in range(self.numParticles):
particleSampleDistribution.append(util.sample(allPossible))
self.particlesList = particleSampleDistribution
def observeState(self, gameState):
"""
Resamples the set of particles using the likelihood of the noisy
observations.
To loop over the ghosts, use:
for i in range(self.numGhosts):
...
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell, position
self.getJailPosition(i) where `i` is the index of the ghost.
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeParticles. After all
particles are generated randomly, any ghosts that are eaten (have
noisyDistance of None) must be changed to the jail Position. This
will involve changing each particle if a ghost has been eaten.
self.getParticleWithGhostInJail is a helper method to edit a specific
particle. Since we store particles as tuples, they must be converted to
a list, edited, and then converted back to a tuple. This is a common
operation when placing a ghost in jail.
"""
pacmanPosition = gameState.getPacmanPosition()
noisyDistances = gameState.getNoisyGhostDistances()
if len(noisyDistances) < self.numGhosts:
return
emissionModels = [busters.getObservationDistribution(dist) for dist in noisyDistances]
"*** YOUR CODE HERE ***"
# counter
possibleCounter = util.Counter()
for _, particle in enumerate(self.particles):
prior = 1
# for each ghost
for i in range(self.numGhosts):
# check the first case for eaten ghost
if noisyDistances[i] is None:
particle = self.getParticleWithGhostInJail(particle, i)
# else calculate the actual distance to get the partial/priority
else:
distance = util.manhattanDistance(pacmanPosition, particle[i])
noisyDistance = emissionModels[i][distance]
prior = prior * noisyDistance
# add prior to particle
possibleCounter[particle] = possibleCounter[particle] + prior
# check for counter values all = 0
if possibleCounter.totalCount() == 0:
self.initializeParticles()
# set particle to jailed ghost because found
for particle in self.particles:
for i in range(self.numGhosts):
if noisyDistances[i] == None:
particle = self.getParticleWithGhostInJail(particle, i)
else:
# else normalize and resample it for particle list
possibleCounter.normalize()
newParticleList = list()
for _ in range(self.numParticles):
newParticleList.append(util.sample(possibleCounter))
self.particles = newParticleList
def observe(self, observation, gameState):
"""
Update beliefs based on the given distance observation. Make sure to
handle the special case where all particles have weight 0 after
reweighting based on observation. If this happens, resample particles
uniformly at random from the set of legal positions
(self.legalPositions).
A correct implementation will handle two special cases:
1) When a ghost is captured by Pacman, all particles should be updated
so that the ghost appears in its prison cell,
self.getJailPosition()
As before, you can check if a ghost has been captured by Pacman by
checking if it has a noisyDistance of None.
2) When all particles receive 0 weight, they should be recreated from
the prior distribution by calling initializeUniformly. The total
weight for a belief distribution can be found by calling totalCount
on a Counter object
util.sample(Counter object) is a helper method to generate a sample from
a belief distribution.
You may also want to use util.manhattanDistance to calculate the
distance between a particle and Pacman's position.
"""
noisyDistance = observation
emissionModel = busters.getObservationDistribution(noisyDistance)
pacmanPosition = gameState.getPacmanPosition()
allPossible = util.Counter()
# 1) if ghost is eaten place it in jail with a probability of 1
if noisyDistance == None:
temp = []
for p in range(len(self.particles)):
temp.append(self.getJailPosition())
self.particles = temp
# otherwise go ahead ahead and create our counter
else:
currentBeliefs = self.getBeliefDistribution()
for p in self.particles:
# find the true distance from pacman to each particle
trueDistance = util.manhattanDistance(p, pacmanPosition)
if emissionModel[trueDistance] > 0:
# weight each particle by the probability of getting to that position (use emission model)
# account for our current belief distribution (evidence) at this point in time
allPossible[
p] = emissionModel[trueDistance] * currentBeliefs[p]
# 2) if all particles have 0 weight, recreate prior distribution
if allPossible.totalCount() == 0:
self.initializeUniformly(gameState)
# otherwise, go ahead and resample based on our new beliefs
else:
keys = []
values = []
# find each key, value pair in our counter
for key, value in allPossible.items():
keys.append(key)
values.append(value)
# resample self.particles
self.particles = util.nSample(values, keys, self.numParticles)
