def inner_execution(envDesc, a, g, ep, e):
env = gym.make(envDesc).env
print("current alpha -> {}, gamma -> {}, epsilon -> {}, episodes -> {}".
format(a, g, ep, e))
qlearn = QLearning(env,
alpha=a,
gamma=g,
epsilon=ep,
epsilon_min=0.001,
epsilon_dec=0.9999,
episodes=e)
q_table = qlearn.train(
"grid_data/q_table_{}_alpha_{}_gamma_{}_ep{}_e{}.csv".format(
envDesc, a, g, ep, e), None)
rewards = 0
for i in range(101):
state = env.reset()
train_done = False
count = 0
while (not train_done) and (count < 200):
action = np.argmax(q_table[state])
state, reward, train_done, _ = env.step(action)
count += 1
if reward == 1:
rewards += 1
r = np.array([a, g, ep, e, rewards])
print(r)
savetxt("grid_results/results_{}_alpha_{}_gamma_{}_ep{}_e{}".format(
envDesc, a, g, ep, e),
r,
delimiter=',',
newline=" ",
fmt="%10.5f")
def inner_execution(env, envDesc, a, g, ep, e):
print("current alpha -> {}, gamma -> {}, epsilon -> {}, episodes -> {}".
format(a, g, ep, e))
qlearn = QLearning(env,
alpha=a,
gamma=g,
epsilon=ep,
epsilon_min=0.001,
epsilon_dec=0.9999,
episodes=e)
q_table = qlearn.train(
"grid_data/q_table_{}_alpha_{}_gamma_{}_ep{}_e{}.csv".format(
envDesc, a, g, ep, e),
"grid_results/actions_{}_alpha_{}_gamma_{}_ep{}_e{}".format(
envDesc, a, g, ep, e))
rewards = 0
for i in range(101):
state = env.reset()
train_done = False
count = 0
while (not train_done) and (count < 200):
action = np.argmax(q_table[state])
state, reward, train_done, _ = env.step(action)
count += 1
if reward == 1:
rewards += 1
self.results.append([a, g, ep, e, rewards])
# Implement Q-learning and use this to solve the cartpole-environment
import gym
# Source: https://github.com/JoeSnow7/Reinforcement-Learning/blob/master/Cartpole%20Q-learning.ipynb
# We define a class to contain the learning algorithm
from QLearning import QLearning
env = gym.make("CartPole-v0")
agent = QLearning(env)
agent.train()
agent.run()
# have a look at LearningPolicy.py for other policies
epsilon_policy = LearningPolicy.exponentially_annealed_epsilon(1 / 10000, 0.0)
epsilon_policy_2 = LearningPolicy.linear_annealed_epsilon(1., 0.1, 100)
alpha1 = 0.2
alpha2 = 0.1
hyperparameters = {"alpha": alpha2, "discount": 0.99}
# Please note: Numerous other settings can be adjusted in settings.py
if training_mode:
q = QLearning(epsilon_policy=epsilon_policy_2,
map_name=map,
hyperparameters=hyperparameters,
save_name=save_name)
while True:
q.train()
else:
q = QLearning(epsilon_policy=LearningPolicy.constant_epsilon(0),
map_name=map)
if checkpoint_file is None:
raise Exception("Please specify the checkpoint file path!")
q_values = AgentManager.load_q_values(checkpoint_file)
while True:
q.test(q_values=q_values)
import matplotlib.pyplot as plt
from QLearning import QLearning
from numpy import loadtxt
def stateNumber(state):
(x,y,z) = state
y = y * 32
z = z * 352
return x+y+z
env = gym.make('Blackjack-v0')
for i in [0.01]:
for g in [0.000001,0.00001,0.0001,0.001,0.01]:
for epi in [600000,700000,800000]:
qlearn = QLearning(env, alpha=i, gamma=g, epsilon=0.9,epsilon_min=0.01, epsilon_dec=0.99, episodes=epi)
q_table = qlearn.train('data/q-table-blackjack.csv', 'results/blackjack')
#q_table = loadtxt('data/q-table-blackjack.csv', delimiter=',')
#state= env.reset()
#print(state)
#state = stateNumber(state)
#done = False
#
#
#while not done:
# action = np.argmax(q_table[state])
# state, reward, done, info = env.step(action)
# print(action)
# print(state)
# state = stateNumber(state)
#
# #print(frame['frame'].getvalue())
# print(f"Timestep: {i + 1}")
# print(f"State: {frame['state']}")
# print(f"Action: {frame['action']}")
# print(f"Reward: {frame['reward']}")
# sleep(.1)
env = gym.make('Roulette-v0').env
#q_table = loadtxt('data/q-table-roulette.csv', delimiter=',')
#2600loss - stable
qlearn = QLearning(env, alpha=0.001, gamma=0.001, epsilon=0.9, epsilon_min=0.001, epsilon_dec=0.9999, episodes=1000000)
# 500-1000loss - real player like
#qlearn = QLearning(env, alpha=0.001, gamma=0.001, epsilon=0.9, epsilon_min=0.1, epsilon_dec=0.7, episodes=1000000)
q_table = qlearn.train('data/q-table-roulette.csv', None)
#q_table = loadtxt('data/q-table-roulette.csv', delimiter=',')
state = env.reset()
done = False
rewards = 0
actions = 0
while not done:
action = np.argmax(q_table)
state, reward, done, info = env.step(action)
actions += 1
rewards += reward
