def test_player_betting():
p = Player('Fooman')
h = Hand()
p.hand_list.append(h)
p.bet(10)
# balance should now be -10 and bet of the hand 10.
assert_true(p.balance == -10 and h.bet == 10)
def test_player_betting():
p = Player('Fooman')
h = Hand()
p.hand_list.append(h)
p.bet(10)
# balance should now be -10 and bet of the hand 10.
assert_true(p.balance == -10 and h.bet == 10)
dealer.show_hand = True
while hand_total(dealer.hand) < 17:
print('*** Dealing card to dealer...')
dealer.get_card(deck.deal())
print(dealer)
player_total = hand_total(player.hand)
dealer_total = hand_total(dealer.hand)
if player_total > 21:
print('*** Player busts...')
player.lose_bet()
elif player_total == dealer_total:
print('*** It is a tie!')
player.bet = 0;
elif player_total > dealer_total:
print('*** Player wins!!!')
player.win_bet()
else:
print('*** Player loses...')
player.lose_bet()
print(dealer)
print(player)
if player.chips == 0:
print('*** Player is out of funds... Game over.')
break;
answer = input('Play again? (y/n) ')
class Agent:
def __init__(self):
self.p = Player()
self.actions = { 1 : self.p.hit,
2 : self.p.stand,
3 : self.p.double,
4 : self.p.split,
}
self.Q = np.empty([0,5],dtype=object)
def choose_action(self,s,epsilon):
possibleActions = np.where(s!=-np.inf)[0][1:]
if (np.random.rand()) <= epsilon:
action = np.random.choice(possibleActions)
else:
action = np.argmax(s[1:]) + 1
return action
def add_state(self,s):
"""
Q-matrix:
0 | 1 | 2 | 3 | 4 |
|(Hit)|(Stand)|(Double)|(Split)|
states |
. |
. |
"""
Qrow = np.hstack((s,np.array((0.,0.,0.,0.),dtype=object)))
if s[1] == "1": # Doubling is not allowed after hitting
Qrow[-2] = -np.inf
else:
Qrow[-2] = np.random.rand(1)[0]*0.1
if s[2] == "0":
Qrow[-1] = -np.inf
else:
Qrow[-1] = np.random.rand(1)[0]*0.1
# Randomly initialize Q's available actions
Qrow[1:3] = np.random.rand(2)*0.1
self.Q = np.vstack((self.Q,Qrow))
def get_reward(self,p):
if p.gameState == "Lost":
return -1.0
elif p.gameState == "Won":
return 1.0
elif p.gameState == "InPlay":
return 0.1
elif p.gameState == "Push":
return 0.5
elif p.gameState == "Natural":
return 0.0
def update_Q(self,Q,p,s,a,ns,na,r,mu,gamma):
if p.gameState != "InPlay":
return mu*(r + gamma*(r - Q[s,a]))
else:
return mu*(r + gamma*Q[ns,na]-Q[s,a])
def learn(self,mu,gamma,epsilon,nGames,plots=False):
x = []
# Episodic epsilon-greedy algorithm
for i in range(nGames):
print("game number %s " % i)
reward = 0
if len(self.p.table.splitHand)==0:
self.p.bet(1)
if self.p.gameState == "Natural": self.p.bet(1)
else:
self.p.table.hand = self.p.table.splitHand.pop()
self.p.table.hand.is_split = False
self.p.gameState = "InPlay"
while self.p.gameState == "InPlay":
stateKey = encode_state(self.p.table.hand,self.p.table.dealer_hand)
# If we haven't seen this state before, extend Q
if stateKey not in self.Q[:,0]:
self.add_state(stateKey)
stateIndex = np.where(stateKey==self.Q[:,0])[0][0]
state = self.Q[stateIndex]
a = self.choose_action(state,epsilon)
self.actions[a]()
nextStateKey = encode_state(self.p.table.hand,self.p.table.dealer_hand)
if nextStateKey not in self.Q[:,0]:
self.add_state(nextStateKey)
nextStateIndex = np.where(nextStateKey==self.Q[:,0])[0][0]
nextState = self.Q[nextStateIndex]
na = self.choose_action(nextState,epsilon)
reward += self.get_reward(self.p)
self.Q[stateIndex,a] += self.update_Q(self.Q,
self.p,
stateIndex,
a,
nextStateIndex,
na,
reward,
mu=mu,
gamma=gamma)
# x.append((1-self.p.losses/(i+1)))*100
yield (1-self.p.losses/(i+1))*100
# print("Wins: ",self.p.wins)
# print("Losses: ", self.p.losses)
# print("Pushes: ",self.p.pushes)
# print("Naturals: ",self.p.naturals)
# pcnt = 1.0 - (self.p.losses/nGames)
# print("Percent not lost: ",pcnt*100,"%")
#
# # if plots:
# plt.plot(x)
# plt.ylabel("Percent won/drawn")
# plt.xlabel("Games played")
# plt.show()
def save(self,filename):
dump(self.Q,open(filename,"wb"))
def load(self,filename):
Q = load(open(filename,"rb"))
self.Q = Q
def play(self,nGames):
# Episodic epsilon-greedy algorithm
for i in range(nGames):
self.p.bet(1)
if self.p.gameState != "InPlay": self.p.bet(1)
while self.p.gameState == "InPlay":
stateKey = encode_state(self.p.table.hand,self.p.table.dealer_hand)
stateIndex = np.where(stateKey==self.Q[:,0])[0][0]
state = self.Q[stateIndex]
a = self.choose_action(state,epsilon=-np.inf)
self.actions[a]()
print("Wins: ",self.p.wins)
print("Losses: ", self.p.losses)
print("Pushes: ",self.p.pushes)
print("Naturals: ",self.p.naturals)
pcnt = 1.0 - (self.p.losses/nGames)
print("Percent not lost: ",pcnt*100,"%")
#Before we start, we need to create deck
mydeck = Deck()
mydeck.shuffle()
#THis will be repeated until player decide to stop playing
while True:
if player1.balance <= 0:
print('Player have not money anymore. Game is finished!')
break
#in case it is not first round we have to discard all cards from hand
player1.discard_hand()
player2.discard_hand()
#Before game each player pick their cards, Human players have to make a bet
player1_bet = player1.bet()
#Then, each player have to pick 2 cards
player1.pick_cards(mydeck)
player2.pick_cards(mydeck)
#Make one of computer cards hidden for player
player2.make_hidden(True)
#Player turn
#Check if player BUST (21 points or more)
#Until player pass 21 points there is still reason to ask him what he want to do
game.player_turn(mydeck, player1, player2)
#computer turn
game.computer_turn(mydeck, player1, player2, player1_bet)
