#sim((run, Prob, P))
print "-> RUN "+str(run)
# count history
#ch = np.zeros(MAXITER+1, dtype=INTTYPE)
# controller initial state
ci = L.names[random.choice(L.S)]
# environment initial state (non observable)
ei = M.names[random.choice(M.S)]
# data tracking
ch[run,0] = func.counter(ci,ei,DIM,COUNTER)
init = { s for s in P.S if P.L.names[s[0]] == ci }
distr = { s:1./len(init) for s in init }
# initialize belief state
b = belief.belief(P,P.Z,init=distr)
# first observation
o = func.weighted_choice(
P.Z[((L.inv_names[ci]),(M.inv_names[ei]))])
# belief update
b.update(obs=o)
if PRINT:
print 'b(s):',b.d[((L.inv_names[ci]),(M.inv_names[ei]))] \
if ((L.inv_names[ci]),(M.inv_names[ei])) in b.d else 0
rob.print_grid(ci,ei,DIM,pause=PAUSE)
for it in range(MAXITER):
# scheduled choice
a = None
if SCHEDULER == 0:
a = sched.beliefScheduler(b, P, Prob, sched.avgMinVal, verbose=PRINT)
elif SCHEDULER == 1:
a = sched.beliefScheduler(b, P, Prob, sched.maxMinVal, verbose=PRINT)
elif SCHEDULER == 2:
Prob = imp.importProb(PRISMFILENAME,NPAR)
for run in range(RUNS):
#sim((run, Prob, P))
print "-> RUN "+str(run)
# controller initial state
ci = L.names[random.choice(L.S)]
# environment initial state (not observable)
ei = 'tl' #M.names[random.choice([M.inv_names['tl'],M.inv_names['tr']])]
init = { s for s in P.S if P.L.names[s[0]] == ci }
distr = {(0,M.inv_names['tl']):0.5, (0,M.inv_names['tr']):0.5}
# initialize belief state
b = belief.belief(P,P.Z,init=distr)
# first observation
o = func.weighted_choice(
P.Z[((L.inv_names[ci]),(M.inv_names[ei]))])
# belief update
print 'init: ', b.d
b.update(obs=o)
if PRINT:
print 'b(s):',b.d[((L.inv_names[ci]),(M.inv_names[ei]))] \
if ((L.inv_names[ci]),(M.inv_names[ei])) in b.d else 0
for it in range(MAXITER):
print 'belief: ', b.d
# scheduled choice
a = None
if SCHEDULER == 0:
a = sched.beliefScheduler(b, P, Prob, sched.avgMinVal, verbose=PRINT)
elif SCHEDULER == 1:
a = sched.beliefScheduler(b, P, Prob, sched.maxMinVal, verbose=PRINT)
maximum = -1
act_max = None
for a in L.A:
ba = b.returnUpdate(act=a)
val = sched.avgMinScheduler(ba,Prob,P)
#val = rob.opt(ba,Prob,P)
if round(val,10) == round(maximum,10):
act_max.add(a)
print 'EQ act:',a,' - val:',val
elif val > maximum:
maximum = val
act_max = {a}
print 'OK act:',a,' - val:',val
else:
print 'NO act:',a,' - val:',val
pass
a = random.choice(list(act_max))
# apply action
ci = rob.step(ci,a,DIM)
ei = (rob.step(ei[0],random.choice(L.A),DIM),
rob.step(ei[1],random.choice(L.A),DIM))
# extract observation
o = func.weighted_choice(P.Z[(L.inv_names[ci],M.inv_names[ei])])
# belief update
b.update(act=a,obs=o)
print 'act:',a
print 'obs:',o
print 'b(s):',b.d[(L.inv_names[ci],M.inv_names[ei])] \
if (L.inv_names[ci],M.inv_names[ei]) in b.d else 0
rob.print_grid(ci,ei,DIM,pause=PAUSE)