if __name__ == '__main__':
random.seed(1)
height = 4
width = 4
initPos = (1,1)
concreteInitPos = (initPos,(0,0),(1,2),(2,1),(3,3))
iterations = 10
# partially observable model
pomdp = robotPomdp(height, width)
# initial states
init = set(map(lambda s: pomdp.inv_states[s], filter(lambda s: s[0] == initPos, pomdp.states.values())))
#init = filter(lambda s: s[0] == initPos, pomdp.states.values())
# initial belief
prior = Distribution(set(pomdp.states.keys()),\
lambda el,dom: Distribution.restrictedUniformPdf(el,dom,init))
#prior = Distribution(set(pomdp.states.keys()), Distribution.uniformPdf)
# prior belief updated to the first observation
#prior = beliefUpdateObservation(pomdp,prior,pomdp.inv_observations[(False,False,False)])
# generate explicit MDP
print 'emdp construction'
emdp = ExplicitMarkovDecisionProcess(pomdp,prior)
print 'end (emdp empty)'
# observation classes
collision = filter(lambda o: o[0], pomdp.observations.values())
notCollision = filter(lambda o: not o[0], pomdp.observations.values())
sensing = filter(lambda o: reduce(lambda a,b: a or b, o), pomdp.observations.values())
anyObs = filter(lambda o: True, pomdp.observations.values())
sensAndNotCollide = [o for o in pomdp.observations.values() if (o in sensing and o in notCollision)]
for (s1,o1),a in it.product(S,pomdp.actions.values()):
for (s2,o2) in S:
if pomdp.inv_states[s2] in pomdp.transitionFunction[(pomdp.inv_states[s1],pomdp.inv_actions[a])] and pomdp.inv_observations[o2] in pomdp.observationFunction[pomdp.inv_states[s2]]:
if ((s1,o1),a) not in T:
T[((s1,o1),a)] = dict()
T[((s1,o1),a)][(s2,o2)] = pomdp.transitionFunction[(pomdp.inv_states[s1],pomdp.inv_actions[a])][pomdp.inv_states[s2]] * pomdp.observationFunction[pomdp.inv_states[s2]][pomdp.inv_observations[o2]]
print 'S:',S,'\nA:',pomdp.actions.values(),'\nT:',T
MarkovDecisionProcess.__init__(self,S,pomdp.actions.values(),T)
# >>> main test
if __name__ == '__main__':
pomdp = cs.tigerPomdp()
print pomdp
prior = Distribution(set(pomdp.states.keys()),lambda el,dom: Distribution.restrictedUniformPdf(el,dom,{pomdp.inv_states['tl'],pomdp.inv_states['tr']}))
# generate explicit MDP
emdp = ExplicitMarkovDecisionProcess(pomdp,prior)
print emdp
# export to DOT
e2d.export2dot(emdp,'emdp','emdp.dot', [], [], [],[])
# >>> authorship information
__author__ = "Marco Tinacci"
__copyright__ = "Copyright 2016"
__credits__ = ["Marco Tinacci"]
__license__ = "ASF"
__version__ = "2.0"
__maintainer__ = "Marco Tinacci"
__email__ = "*****@*****.**"
