def main():
learning_rate = 0.05
discount = 0.9
iterations = 10000
agent = Gambler(learning_rate=learning_rate,
discount=discount,
iterations=iterations)
# setup simulation
mouseGame = MouseGame()
mouseGame.reset()
total_reward = 0 # Score keeping
last_total = 0
# main loop
for step in range(iterations):
old_state = list(mouseGame.mouse) # Store current state
action = agent.get_next_action(
old_state) # Query agent for the next action
new_state, reward = mouseGame.take_action(
action) # Take action, get new state and reward
agent.update(old_state, new_state, action.value,
reward) # Let the agent update internals
total_reward += reward # Keep score
if step % 250 == 0: # Print out metadata every 250th iteration
performance = (total_reward - last_total) / 250.0
print({
'step': step,
'performance': performance,
'total_reward': total_reward
})
last_total = total_reward
time.sleep(0.00001) # Avoid spamming stdout too fast!
# print("Final Q-table", agent.q_table)
for i in range(len(agent.q_table)):
for j in range(len(agent.q_table[i])):
#print("[" + str(i) + "][" + str(j) + "]: ", end="")
#print(agent.q_table[i][j])
print("[%d][%d]:" % (i, j), agent.q_table[i][j])
input() # so console window doesnt close on windows
def martingale(initialWealth, betAmount):
roulette = Roulette()
gambler = Gambler(initialWealth)
while gambler.getWealth() >= betAmount:
winnings = roulette.betColor(betAmount)
gambler.decrementWealth(betAmount)
gambler.incrementWealth(winnings)
if winnings > 0:
break
betAmount *= 2
return gambler.getWealth()
def main():
# parse arguments
parser = argparse.ArgumentParser()
parser.add_argument('--agent', type=str, default='GAMBLER', help='Which agent to use')
parser.add_argument('--learning_rate', type=float, default=0.1, help='How quickly the algorithm tries to learn')
parser.add_argument('--discount', type=float, default=0.95, help='Discount for estimated future action')
parser.add_argument('--iterations', type=int, default=2000, help='Iteration count')
FLAGS, unparsed = parser.parse_known_args()
# select agent
if FLAGS.agent == 'GAMBLER':
agent = Gambler(learning_rate=FLAGS.learning_rate, discount=FLAGS.discount, iterations=FLAGS.iterations)
elif FLAGS.agent == 'ACCOUNTANT':
agent = Accountant()
elif FLAGS.agent == 'DEEPGAMBLER':
agent = DeepGambler(learning_rate=FLAGS.learning_rate, discount=FLAGS.discount, iterations=FLAGS.iterations)
else:
agent = Drunkard()
# setup simulation
dungeon = DungeonSimulator()
dungeon.reset()
total_reward = 0 # Score keeping
last_total = 0
# main loop
for step in range(FLAGS.iterations):
old_state = dungeon.state # Store current state
action = agent.get_next_action(old_state) # Query agent for the next action
new_state, reward = dungeon.take_action(action) # Take action, get new state and reward
agent.update(old_state, new_state, action, reward) # Let the agent update internals
total_reward += reward # Keep score
if step % 250 == 0: # Print out metadata every 250th iteration
performance = (total_reward - last_total) / 250.0
print(json.dumps({'step': step, 'performance': performance, 'total_reward': total_reward}))
last_total = total_reward
time.sleep(0.0001) # Avoid spamming stdout too fast!
def main():
print("xxx")
parser=argparse.ArgumentParser()
parser.add_argument("--agent", type=str, default = 'DEEPGAMBLER',help="Which agent to use")
parser.add_argument("--learning_agent", type=float, default = 0.1,help="Choose Learning Rate ")
parser.add_argument("--discount", type=float, default = 0.95, help="choose discount")
parser.add_argument("--iterations", type=int, default = 1000,help="No of iterations")
FLAGS,unparsed = parser.parse_known_args()
print("xxx")
if FLAGS.agent == "DRUNKARD":
agent= Drunkard()
elif FLAGS.agent=="ACCOUNTANT":
agent=Accountant()
elif FLAGS.agent == "GAMBLER":
agent=Gambler(FLAGS.learning_agent,FLAGS.discount,1.0,FLAGS.iterations)
else:
agent=DeepGambler(FLAGS.learning_agent,FLAGS.discount,1.0,FLAGS.iterations)
dungeon= DungeonSimulator()
dungeon.reset()
total_reward = 0
print("agent created")
for step in range(FLAGS.iterations):
old_state = dungeon.state
action = agent.get_next_action(old_state)
new_state , reward = dungeon.take_action(action)
agent.update(old_state, new_state, action, reward)
total_reward += reward
if step %250 ==0:
print(json.dumps({'step': step,'total-reward':total_reward, }))
time.sleep(0.0001)
print("FINAL Q TABLE", agent.q_table)
def example_1():
"""
Example 1: Obtains the solution for a given gambler problem using value
iteration and policy iteration.
"""
# Initialises all required inputs for the Gambler.
name = 'base_problem'
goal = 100
win_prob = 0.4
discount = 0.9
# Initialises the Gambler.
gambler = Gambler(name, goal, win_prob, discount)
# Obtains the optimum value estimate and policy from the Gambler using
# value iteration.
gambler.obtain_optimum('value_iteration')
# Obtains the optimum value estimate and policy from the Gambler using
# policy iteration.
gambler.obtain_optimum('policy_iteration')
from blackjack import Blackjack
from gambler import Gambler
from card import Suit, CardFace, Card
from random import shuffle
# spawns a blackjack game with 7 decks.
deck = Blackjack(7)
player = Gambler()
dealer = Gambler()
playing = True
def print_dealt_card(self, card_suit, card_value):
print("cards")
def deal_player_card(self):
for hand in player.hand:
hand.append(deck.deal_card)
while (playing):
print("Enter S to start a new game, or Q to quit")
main_selection = raw_input()
main_selection = main_selection.lower()
dealing = None
if (main_selection == 's'):
dealing = True