def reinforce_learning():
Qsa, Csa = {}, {}
for s1 in PLAYER_STATES:
# s1 means sum of player's cards
for s2 in DEALER_STATES:
# s2 means the first card of dealer
# about action
# k=0 -> Hit, k=1 -> Stick
Qsa[(s1, s2)] = [0, 0]
Csa[(s1, s2)] = False
while True:
s1, s2 = choice(PLAYER_STATES), choice(DEALER_STATES)
for a in xrange(2):
if a == 0:
# HIT
new_card = get_card()
if new_card[0] == RED:
new_s1 = s1 - new_card[1]
else:
assert new_card[0] == BLACK
new_s1 = s1 + new_card[1]
if new_s1 < 1 or new_s1 > 21:
qrs = LOSE
elif new_s1 == 21:
qrs = WIN
else:
new_Q = []
for k, v in Qsa.iteritems():
_s1, _s2 = k
if _s1 == new_s1:
new_Q += v
qrs = 0 + GAMMA * max(new_Q)
elif a == 1:
# STICK
# Dealer's turn
new_s2 = s2
while True:
if new_s2 < 1 or new_s2 > 21:
qrs = WIN
break
elif new_s2 >= 17:
if s1 == new_s2:
qrs = DRAW
elif s1 > new_s2:
qrs = WIN
else:
assert s1 < new_s2
qrs = LOSE
break
else:
# HIT
new_card = get_card()
if new_card[0] == RED:
new_s2 -= new_card[1]
else:
assert new_card[0] == BLACK
new_s2 += new_card[1]
else:
assert False
Qsa[(s1, s2)][a] = (1 - ALPHA) * Qsa[(s1, s2)][a] + ALPHA * qrs
# Check where the state is converged or not
if abs(Qsa[(s1, s2)][0] - Qsa[(s1, s2)][1]) > CONVERGENCE_CONDITION:
Csa[(s1, s2)] = True
# If all states are converged at once, end reinforce learning
for v in Csa.itervalues():
if not v:
break
else:
break
return Qsa
def play_round(way_to_decision):
# Initialize a game
player_cards, dealer_cards = [], []
while True:
new_card = get_card()
if new_card[0] == RED:
continue
else:
break
player_cards.append(new_card)
while True:
new_card = get_card()
if new_card[0] == RED:
continue
else:
break
dealer_cards.append(new_card)
#
rv_player = None
while True:
if DISPLAY:
print "Player's turn"
display_current(player_cards, dealer_cards)
player_choice = way_to_decision(player_cards, dealer_cards)
if player_choice == HIT:
nc = get_card()
if DISPLAY:
print 'New card is (%s, %d)' % ('Red' if nc[0] == RED else
'Black', nc[1])
player_cards.append(nc)
s = get_sum(player_cards)
if s < 1 or s > 21:
rv_player = BUST
break
if s == 21:
return WIN
else:
assert player_choice == STICK, player_choice
rv_player = STICK
break
#
if rv_player != BUST:
# the player did not go bust
rv_dealer = None
while True:
if DISPLAY:
print "Dealer's turn"
display_current(player_cards, dealer_cards)
s = get_sum(dealer_cards)
if s < 1 or s > 21:
rv_dealer = BUST
break
elif s >= 17:
rv_dealer = STICK
break
if s == 21:
return LOSE
else:
# HIT
nc = get_card()
if DISPLAY:
print 'New card is (%s, %d)' % ('Red' if nc[0] == RED else
'Black', nc[1])
dealer_cards.append(nc)
if rv_dealer == BUST:
return WIN
player_sum, dealer_sum = get_sum(player_cards), get_sum(dealer_cards)
if player_sum == dealer_sum:
return DRAW
return WIN if player_sum > dealer_sum else LOSE
def play_round(way_to_decision):
# Initialize a game
player_cards, dealer_cards = [], []
while True:
new_card = get_card()
if new_card[0] == RED:
continue
else:
break
player_cards.append(new_card)
while True:
new_card = get_card()
if new_card[0] == RED:
continue
else:
break
dealer_cards.append(new_card)
#
rv_player = None
while True:
if DISPLAY:
print "Player's turn"
display_current(player_cards, dealer_cards)
player_choice = way_to_decision(player_cards, dealer_cards)
if player_choice == HIT:
nc = get_card()
if DISPLAY:
print 'New card is (%s, %d)' %('Red' if nc[0]==RED else 'Black', nc[1])
player_cards.append(nc)
s = get_sum(player_cards)
if s < 1 or s > 21:
rv_player = BUST
break
if s == 21:
return WIN
else:
assert player_choice == STICK, player_choice
rv_player = STICK
break
#
if rv_player != BUST:
# the player did not go bust
rv_dealer = None
while True:
if DISPLAY:
print "Dealer's turn"
display_current(player_cards, dealer_cards)
s = get_sum(dealer_cards)
if s < 1 or s > 21:
rv_dealer = BUST
break
elif s >= 17:
rv_dealer = STICK
break
if s == 21:
return LOSE
else:
# HIT
nc = get_card()
if DISPLAY:
print 'New card is (%s, %d)' %('Red' if nc[0]==RED else 'Black', nc[1])
dealer_cards.append(nc)
if rv_dealer == BUST:
return WIN
player_sum, dealer_sum = get_sum(player_cards), get_sum(dealer_cards)
if player_sum == dealer_sum:
return DRAW
return WIN if player_sum > dealer_sum else LOSE
def reinforce_learning():
Qsa, Csa = {}, {}
for s1 in PLAYER_STATES:
# s1 means sum of player's cards
for s2 in DEALER_STATES:
# s2 means the first card of dealer
# about action
# k=0 -> Hit, k=1 -> Stick
Qsa[(s1, s2)] = [0, 0]
Csa[(s1, s2)] = False
while True:
s1, s2 = choice(PLAYER_STATES), choice(DEALER_STATES)
for a in xrange(2):
if a == 0:
# HIT
new_card = get_card()
if new_card[0] == RED:
new_s1 = s1 - new_card[1]
else:
assert new_card[0] == BLACK
new_s1 = s1 + new_card[1]
if new_s1 < 1 or new_s1 > 21:
qrs = LOSE
elif new_s1 == 21:
qrs = WIN
else:
new_Q = []
for k, v in Qsa.iteritems():
_s1, _s2 = k
if _s1 == new_s1:
new_Q += v
qrs = 0 + GAMMA * max(new_Q)
elif a == 1:
# STICK
# Dealer's turn
new_s2 = s2
while True:
if new_s2 < 1 or new_s2 > 21:
qrs = WIN
break
elif new_s2 >= 17:
if s1 == new_s2:
qrs = DRAW
elif s1 > new_s2:
qrs = WIN
else:
assert s1 < new_s2
qrs = LOSE
break
else:
# HIT
new_card = get_card()
if new_card[0] == RED:
new_s2 -= new_card[1]
else:
assert new_card[0] == BLACK
new_s2 += new_card[1]
else:
assert False
Qsa[(s1, s2)][a] = (1 - ALPHA) * Qsa[(s1, s2)][a] + ALPHA * qrs
# Check where the state is converged or not
if abs(Qsa[(s1, s2)][0] - Qsa[(s1, s2)][1]) > CONVERGENCE_CONDITION:
Csa[(s1, s2)] = True
# If all states are converged at once, end reinforce learning
for v in Csa.itervalues():
if not v:
break
else:
break
return Qsa
