T = {}
for s1,a in it.product(MS,L.A):
# TOFIX
if NROB == 2:
S2 = rob.around(s1,DIM)
else: # NROB > 2
S2 = rob.around(s1[0],DIM)
for i in range(1,NROB-1):
S2 = list(it.product(S2,rob.around(s1[i],DIM)))
p = 1.0 / float(len(S2))
T[(s1,a)] = {}
for s2 in S2:
s = (s2[0],s2[1])
T[(s1,a)][s] = p
M = mdp.mdp(MS,L.A,T)
# ==== POMDP partially observable ====
P = pomdp.pomdp()
P.initProduct(L,M)
# single observation
single_obs = ['n','s','e','w','h']
# collective observation
P.O = list(func.powerset(single_obs))
# inexpensive state space construction
P.Z = {}
for (l,m) in P.S:
# print l,m
# if NROB > 2:
MT = {
('tl','l') : { 'tl' : 1 },
('tl','ol') : { 'lose' : 1 },
('tl','or') : { 'win' : 1 },
('tr','l') : { 'tr' : 1 },
('tr','ol') : { 'win' : 1 },
('tr','or') : { 'lose' : 1 },
('win','l') : { 'win' : 1 },
('win','ol') : { 'win' : 1 },
('win','or') : { 'win' : 1 },
('lose','l') : { 'lose' : 1 },
('lose','ol') : { 'lose' : 1 },
('lose','or') : { 'lose' : 1 }
}
M = mdp.mdp(MS,L.A,MT)
# ==== POMDP partially observable ====
P = pomdp.pomdp()
P.initProduct(L,M)
# observations
P.O = ['hl','hr']
# observation function
P.Z = {
(0,P.M.inv_names['tl']) : {'hl' : 0.85, 'hr' : 0.15},
(0,P.M.inv_names['tr']) : {'hl' : 0.15, 'hr' : 0.85},
(0,P.M.inv_names['win']) : {'hl' : 0.5, 'hr' : 0.5},
(0,P.M.inv_names['lose']) : {'hl' : 0.5, 'hr' : 0.5}
}
T = {}
for s1,a in it.product(MS,L.A):
# TOFIX
if NROB == 2:
S2 = rob.around(s1,DIM)
else: # NROB > 2
S2 = rob.around(s1[0],DIM)
for i in range(1,NROB-1):
S2 = list(it.product(S2,rob.around(s1[i],DIM)))
p = 1.0 / float(len(S2))
T[(s1,a)] = {}
for s2 in S2:
s = (s2[0],s2[1])
T[(s1,a)][s] = p
M = mdp.mdp(MS,L.A,T)
# ==== POMDP partially observable ====
print "-> POMDP"
print "--> INIT PRODUCT POMDP"
P = pomdp.pomdp()
P.initProduct(L,M)
print "--> OBSERVATION FUNCTION"
P.O = ['some','none']
# inexpensive state space construction
P.Z = {}
for (l,m) in P.S:
#print l,m
if NROB > 2: # TOFIX