action = RL.choose_action(observation)
observation_,reward,done = env.step(action)
RL.store_transition(observation, action, reward, observation_)
if (step > 200) and (step % 5 == 0):
RL.learn()
# swap observation
observation = observation_
# break while loop when end of this episode
if done:
break
step += 1
print('game over')
env.destroy()
if __name__ == '__main__':
env = Maze()
RL = DoubleDQN(env.n_actions, env.n_features,
learning_rate=0.01,
reward_decay=0.9,
e_greedy=0.9,
replace_target_iter=200,
memory_size=2000,
# output_graph=True
)
env.after(100, run_maze)
env.mainloop()
RL.plot_cost()
from maze_env import Maze
from RL_brain import Agent
from RL_brain import myAgent
import time
import matplotlib as plt
from pylab import *
if __name__ == "__main__":
### START CODE HERE ###
# This is an agent with random policy. You can learn how to interact with the environment through the code below.
# Then you can delete it and write your own code.
graph_episode = []
graph_reward = []
env = Maze()
agent = myAgent(actions=list(range(env.n_actions)))
for episode in range(500):
s = env.reset()
episode_reward = 0
agent.epoch_num += 1
while True:
#env.render() # You can comment all render() to turn off the graphical interface in training process to accelerate your code.
a = agent.choose_action(s)
s_, r, done = env.step(a)
q_dict = agent.update(s, a, s_, r)
episode_reward += r
s = s_
agent.has_been_to_this_state[float((s[0] + s[2]) / 2),
float((s[1] + s[3]) / 2)] = True
if done:
#env.render()
#time.sleep(0.5)
# swap observation
observation = observation_
# break while loop when end of this episode
if done:
break
step += 1
# end of game
print('run over')
time_elapsed = time.time() - since
# 代码计时
print('The run_maze code run {:.0f}m {:.0f}s'.format(
time_elapsed // 60, time_elapsed % 60))
if __name__ == "__main__":
env = Maze()
RL = DeepQNetwork(env.n_actions, env.n_features,
learning_rate=0.01,
reward_decay=0.9,
e_greedy=0.9,
replace_target_iter=200,
memory_size=2000,
# output_graph=True
)
run_maze()
RL.plot_cost()
from RL_brain import DeepQNetwork
from maze_env import Maze
import tensorflow as tf
from tqdm import tqdm
import numpy as np
from knn_reward_model import KNN_Predict
MEMORY_SIZE = 20000
epsilon = 200
sess = tf.Session()
np.random.seed(1)
tf.set_random_seed(1)
env = Maze()
with tf.variable_scope('natural_DQN'):
RL_natural = DeepQNetwork(n_actions=env.n_actions,
n_features=env.n_features,
memory_size=MEMORY_SIZE,
replace_target_iter=200
# e_greedy_increment=0.0001
)
knn = KNN_Predict(3)
times = []
real = []
def onehot(action):
x = [0, 0, 0, 0]
x[action] = 1
return x
def plot_reward_movements():
plt.figure()
plt.subplot(2, 1, 1)
plt.plot(episodes, movements)
plt.xlabel("Episode")
plt.ylabel("# Movements")
plt.subplot(2, 1, 2)
plt.step(episodes, rewards)
plt.xlabel("Episode")
plt.ylabel("Reward")
plt.savefig("rewards_movements_q_learn.png")
plt.show()
if __name__ == "__main__":
# Craete maze environment
env = Maze() #ToDo: instanciate Maze class
# Create Q-learning agent
q_learning_agent = QLearningTable(actions=list(range(
env.n_actions))) #ToDo: instanciate QLearningTable class
# Call run_experiment() function once after given time in milliseconds.
env.window.after(10, run_experiment)
# The infinite loop used to run the application, wait for an event to occur and process the event
# till the window is not closed.
env.window.mainloop()
from maze_env import Maze
from RL_brain import QL
import time
# Parameters
EPSILON = 0.9 # Greedy Policy
ALPHA = 0.1 # Learing Rate
LAMBDA = 0.9 # Discount Factor
MAX_EPISODE = 50
MAZE_SIZE = 5
TRAP_SET = [[0, 1], [2, 2], [3, 3], [3, 0]]
TREASURE_SET = [[4, 4]]
env = Maze(MAZE_SIZE)
RL = QL(MAZE_SIZE, EPSILON, LAMBDA, ALPHA)
def update():
for episode in range(MAX_EPISODE):
O = env.reset() # return initial observation
step_number = 0
env.set_trap(TRAP_SET)
env.set_treasure(TREASURE_SET)
is_terminated = False
env.render()
while not is_terminated:
A = RL.choose_action(O)
# print(O)
OO, R, is_terminated = env.step(A)
# print(O,OO)
print('failed!')
env.destroy()
else:
observation = env.reset()
def on_release(key):
print('{0} release'.format(key))
if key == Key.esc:
# Stop listener
return False
def get_action():
global observation
observation = env.reset()
listener = Listener(on_press=on_press, on_release=on_release)
listener.start()
if __name__ == "__main__":
# maze game
env = Maze()
env.after(100, get_action)
print('env started!')
env.mainloop()
print('env ended!')
# Collect events until released
class Controller:
def __init__(self):
self.env = Maze(self)
self.env.mainloop()
def main(self, tubarao, foca, peixe, alga, calorias):
qtd_tub = int(tubarao)
qtd_foca = int(foca)
qtd_peixe = int(peixe)
qtd_alga = int(alga)
cal = int(calorias)
threads = []
MAX_THREAD = qtd_foca + qtd_peixe + qtd_tub + qtd_alga
nome = None
self.env.p.set(MAX_THREAD)
semaforo = threading.Semaphore(MAX_THREAD)
semaforo_validation = threading.Semaphore(0)
teste = threading.Lock()
thread_timer = TimerThread(596, self.env)
thread_timer.start()
next_id = 0
#faz o numero de tubarao
for i in range(qtd_tub):
tipo = 4
thread = IndividuoThread(next_id,
cal,
nome,
tipo,
semaforo_validation,
self.env,
semaforo,
teste=teste,
max_t=MAX_THREAD)
thread.start()
threads.append(thread)
next_id += 1
# faz o numero de foca
for i in range(qtd_foca):
tipo = 3
thread = IndividuoThread(next_id,
cal,
nome,
tipo,
semaforo_validation,
self.env,
semaforo,
teste=teste,
max_t=MAX_THREAD)
thread.start()
threads.append(thread)
next_id += 1
# faz o numero de peixe
for i in range(qtd_peixe):
tipo = 2
thread = IndividuoThread(next_id,
cal,
nome,
tipo,
semaforo_validation,
self.env,
semaforo,
teste=teste,
max_t=MAX_THREAD)
thread.start()
threads.append(thread)
next_id += 1
# faz o numero de peixe
for i in range(qtd_alga):
tipo = 1
thread = IndividuoThread(next_id,
cal,
nome,
tipo,
semaforo_validation,
self.env,
semaforo,
teste=teste,
max_t=MAX_THREAD)
thread.start()
threads.append(thread)
next_id += 1
thread_validacao = ValidationThread(133, threads, semaforo_validation,
semaforo, MAX_THREAD, self.env)
thread_validacao.start()
def __init__(self):
self.env = Maze(self)
self.env.mainloop()
def main():
env = Maze()
agent = QLearning(action_space=list(range(env.n_action)))
update(agent, env)
env.mainloop()
break
step += 1
# end of game
print('game over')
env.destroy()
if __name__ == "__main__":
time_string = utils.get_string_time()
print(time_string, " the test begins")
start_time = time.clock()
# maze game
env = Maze()
RL = DeepQNetwork(
env.n_actions,
env.n_features,
learning_rate=0.01,
reward_decay=0.9,
e_greedy=0.98, # 最终不再探索
replace_target_iter=300,
memory_size=4800,
e_greedy_origin=0.5,
e_greedy_increment=0.0001, # epsilon增长速度
model_load=True,
model_load_dir="save/2018-3-30-22:17/model.ckpt",
model_save_dir="save/{time}/model.ckpt".format(time=time_string),
output_graph=False,
break
step += 1
# end of game
print('game over')
env.destroy()
if __name__ == "__main__":
time_string = utils.get_string_time()
print(time_string, " the test begins")
start_time = time.clock()
# maze game
env = Maze()
RL = DeepQNetwork(
env.n_actions,
env.n_features,
learning_rate=0.02,
reward_decay=0.9,
e_greedy=0.98, # 最终不再探索
replace_target_iter=300,
memory_size=60000,
e_greedy_origin=0.7,
e_greedy_increment=0.00005, # epsilon增长速度
model_load=False,
model_load_dir="save/2018-4-4-16:50/model.ckpt",
model_save_dir="save/{time}/model.ckpt".format(time=time_string),
output_graph=True,
def modify_path(routeX, routeY):
isFind, i1, i2 = find_same(routeX, routeY)
while isFind:
routeX = np.append(routeX[:i1], routeX[i2:])
routeY = np.append(routeY[:i1], routeY[i2:])
#print(routeX)
#print(routeY)
isFind, i1, i2 = find_same(routeX, routeY)
return routeX, routeY
if __name__ == "__main__":
# maze game
env = Maze()
model_path = os.path.dirname(
os.path.abspath(__file__)) + "/model20171219.ckpt"
RL = DeepQNetwork(
env.n_actions,
env.n_features,
model_path,
learning_rate=0.01,
reward_decay=0.9,
e_greedy=0.9,
replace_target_iter=200,
memory_size=2000,
# output_graph=True
)
env.after(100, get_route_maze)
env.mainloop()
print("新观测值，即下一位置：\n", observation_)
print("奖励：\n", reward)
print("done？：\n", done)
print("Q表：\n", RL.q_table)
print("-" * 50 + "\n")
observation = observation_ # 将新观测值作为下一次的初始观测值
# 若下地狱或上天堂，则结束循环
if done:
break
print("~" * 50)
print("第%d回合结束" % (episode + 1))
print("~" * 50 + "\n")
# 游戏结束，销毁环境
print('Game Over')
env.destroy()
RL.q_table.to_csv(
'./model.csv'
) # 【可选】将最终的Q表存储下来，Q表即训练之后的模型；（本想将模型保存下来以便下次直接使用，但pandas的存储与读取遇到小问题，非重点，故暂搁置）
if __name__ == "__main__":
env = Maze() # 创建迷宫环境
RL = QLearningTable(actions=list(range(
env.n_actions))) # 声明RL，即强化学习的行动者，暂称机器人；参数actions=[0, 1, 2, 3]
env.after(100, update) # tkinter的after函数；每100ms调用一次update函数
env.mainloop() # 启动tkinter，即以窗口形式显示环境
while True:
# fresh env
env.render()
# RL choose action based on observation
action = RL.choose_action(str(observation)) #选择
# RL take action and get next observation and reward
observation_, reward, done = env.step(action)
# RL learn from this transition
RL.learn(str(observation), action, reward, str(observation_))
# swap observation
observation = observation_
# break while loop when end of this episode
if done:
break
# end of game
print('game over')
env.destroy()
if __name__ == "__main__":
env = Maze() #先定义一个环境
RL = QLearningTable(actions=list(range(env.n_actions))) #给出一个动作list
env.after(100, update)
env.mainloop()
from maze_env import Maze
from RL_brain import Agent
import time
# from uilts import view
actions = ["up", "down", "right", "left"]
if __name__ == "__main__":
### START CODE HERE ###
# This is an agent with random policy. You can learn how to interact with the environment through the code below.
# Then you can delete it and write your own code.
env = Maze()
agent = Agent(actions=list(range(env.n_actions)))
for episode in range(50):
s = env.reset()
episode_reward = 0
while True:
env.render(
) # You can comment all render() to turn off the graphical interface in training process to accelerate your code.
# time.sleep(0.1)
# view(agent.q_table)
a = agent.choose_action(s)
s_, r, done = env.step(a)
agent.add_observation(s, a, r, s_, done)
episode_reward += r
s = s_
RL.learn(str(observation), action, reward, str(observation_))
record(str(observation), action, reward, str(observation_),
done, RL.q_table, int(int(episode / 10000 + 1) * 10000))
observation = copy.deepcopy(observation_)
if done:
break
def trainWithTrajectory(trajectory, episode):
for item in trajectory:
observation, action, reward, observation_ = item
RL.learn(observation, action, reward, observation)
record(observation, action, reward, observation_, True, RL.q_table,
int(int(episode / 10000 + 1) * 10000))
if __name__ == '__main__':
task_num = [
3,
]
for taskNum in task_num:
parameter["taskNum"] = taskNum
from task import *
task = createTask()
# Q-learning
env = Maze(task)
RL = QLearningTable(actions=list(range(env.n_actions)))
update1(env, RL, 5)
RL.q_table.to_csv("Q_learning Table" + str(taskNum) + "_Insert")
(actor_loss + critic_loss).backward()
self.actor_critic.optimizer.step()
from maze_env import Maze
import os
import time
import matplotlib.pyplot as plt
import pickle
np.random.seed(42)
if __name__ == '__main__':
env = Maze(height=21,
width=21,
detection_range=1,
obstacle_occupancy_prob=0.5)
agent = ACAgent(alpha=0.00001,
gamma=0.999,
input_dims=[529],
n_actions=8,
layer1_size=2048,
layer2_size=512)
score_history = []
avg_score_history = []
epsiodes = 5000000
max_steps = 23 * 23
# RL learn from this transition
RL.learn(str(observation), action, reward, str(observation_),
_action)
action = _action
# swap observation
observation = observation_
# break while loop when end of this episode
if done:
break
# end of game
print('game over')
env.destroy()
if __name__ == "__main__":
env = Maze()
# RL = QLearningTable(actions=list(range(env.n_actions)))
# RL = SarsaTable(actions=list(range(env.n_actions)))
RL = SarsaLambdaTable(actions=list(range(env.n_actions)))
# delay 一段時間後執行cb function
# env.after(100, update)
# update()
update_sarsa()
print(RL.q_table)
env.mainloop()
rew.append(R)
RL.learn(str(stade), action, reward, str(observation_))
# 将下一个 state 的值传到下一次循环
stade = copy.deepcopy(observation_)
# 如果掉下地狱或者升上天堂, 这回合就结束了
if done:
break
plt.plot(np.arange(len(rew)), rew)
plt.ylabel('Reward', fontsize=14)
plt.xlabel('Training steps', fontsize=14)
plt.show()
if __name__ == "__main__":
# 定义环境 env 和 RL 方式
env = Maze()
RL = QLearningTable(actions=list(range(env.n_actions)))
# 开始可视化环境 env
update()
for edgeuser in [5, 10, 15, 20, 25, 30, 35, 40, 45]:
# 随机处理
t = 0
for i in range(10):
G = topology() # 任务图
E1 = E
t += Baseline(user, G, E1, KE, edgeuser)
random.append(t / 10)
# 穷举法
H = topology() # 任务图
E2 = E
# end of game
print('game over for QV-learning')
env.destroy()
if __name__ == "__main__":
plot_y = list(range(EPIS))
fig, ax = plt.subplots()
# Data for plotting
ax.set(xlabel='episode (s)',
ylabel='Total Rewards',
title='Total rewards at each episode')
ax.grid()
#QV-learning
env = Maze()
RL = QVLearningTable(actions=list(range(env.n_actions)),
learning_rate=0.7,
lr_v=0.5)
env.after(100, update_QV_learning)
env.mainloop()
ax.plot(range(EPIS), plot_y, label='QV-learning')
#Q-learning
env = Maze()
RL = QLearningTable(actions=list(range(env.n_actions)),
learning_rate=0.5,
e_greedy=0.85)
env.after(100, update_Q_learning)
env.mainloop()
ax.plot(range(EPIS), plot_y, label='Q-learning')
s1, reward, done = env.step(a0)
a1 = agent.get_action(s1, episode)
agent.learning(s0, a0, reward, s1, a1)
s0 = s1
if done:
if reward < 0:
print("\033[0;31;40m\t{}\033[0m".format(path))
fail += 1
else:
print("\033[0;32;40m\t{}\033[0m".format(path))
success += 1
break
print('game over')
print('success:', success)
print('fail:', fail)
env.destroy()
if __name__ == "__main__":
env = Maze()
# agent = SARSA()
# agent = SARSA_LAMBDA()
# env.after(100, on_policy)
agent = Q_learning()
env.after(100, q_learning)
env.mainloop()
episode = max(plot_episode)
env.destroy()
env.mainloop()
plot_episode.clear()
# This step is a very special bug. We import this function and run this in a for loop.
# The data in plot_episode will be stored in thr for loop.
# We must clear it at the end.
return running_time, episode, plot_sum_reward
# This 'if' judgement is very special, the function is: if we import this file in other places,
# Rather than running this file directly, the content below will not be processed.
if __name__ == "__main__":
print('-----------Start-------------')
env = Maze(height=10, width=10)
Sarsa_brain_ = Model.SARSA(greedy_rate=0.9, learning_rate=0.01, reward_decay=0.9)
# Use two methods to evaluate the algorithm,
# the time it takes to complete the appointed number of episode
# the step it takes to converge (judged by the repeated steps: if the reward in 10 sequent episodes are the same
# We roughly think the algorithm started to converge
# When the judge_method == numbers of repeated steps, the
# update(judge_number=100, judge_method='numbers of repeated steps')
T1 = time.perf_counter()
# update(judge_number=1000, judge_method='sum of episodes', delay_time=0.00)
update(judge_number=10, judge_method='repeated steps', delay_time=0.00)
T2 = time.perf_counter()
print('Time spend :%s ms' % ((T2 - T1)*1000))
result_display.result_plot(x=plot_episode, y=plot_sum_reward, x_name='Episode', y_name='Sum of Reward',
# swap observation
observation = observation_
# break while loop when end of this episode
if done:
break
step += 1
# end of game
print('game over')
env.destroy()
if __name__ == "__main__":
# maze game
env = Maze()
env = gym.make('DoomDefendLine-v0')
env.reset()
RL = DeepQNetwork(2, 4,
learning_rate=0.03,
reward_decay=0.7,
e_greedy=0.9,
replace_target_iter=200,
memory_size=2000,
e_greedy_increment = 0.001
# output_graph=True
)
run_maze()
import os from maze_env import Maze import time import numpy as np import random maz = Maze() # only for interaction: # maz.render() # initiate matrix with dimension from maze dim = maz.MAZE_Limit np.set_printoptions(suppress=True) # action options up down left right a = [(0, 1), (0, -1), (1, 0), (-1, 0)] # crate environment state_value = np.tile(0, dim) counter = np.tile(0, dim) # [DONE] First or Every visit state value counted visit = "first" # "first" or "every" # [DONE] Discounted factor gamma = 0.9 # [DONE] MC # [TO-DO] if there's block # [TO-DO] if random start n_episode = 10000 for n in range(0, n_episode):
main function
"""
#next_stat, reward, done, stat, action = q_learning.observation
#Q = q_learning.Q
##
if __name__ == "__main__":
N_episode = 50
env = Maze()
q_learning = table_qlearning(
n_actions = 4,
epsilon=0.8,
)
for episode in range(N_episode):
print("starting an new episode\n")
stat = env.reset()
i_step = 0
R =[]
while True:
i_step += 1
env.render()
) #从200幕后，每5幕进行一次学习（5就是学习频率learning_freq），更新evaluate-Q网络参数，每200次学习更新一次目标Q网络参数，学习的大脑中的记忆是最近2000幕，也就是可能跨越过程选batch
# swap observation
observation = observation_
# break while loop when end of this episode
if done:
break
step += 1 #每进行一幕，step+1
# end of game
print('game over')
env.destroy()
if __name__ == "__main__":
# maze game
env = Maze()
RL = DeepQNetwork(
env.n_actions,
env.n_features,
learning_rate=0.01,
reward_decay=0.9,
e_greedy=0.9,
replace_target_iter=200, #每200次学习更新一次target-Q参数
memory_size=2000, #记忆容量提到2000，即在最近2000幕中随机选batch
# output_graph=True
)
env.after(100, run_maze)
env.mainloop()
RL.plot_cost()
print(episode, 'episode,', episode_step, 'step_counter, 失败结束')
elif reward == 1:
print(episode, 'episode,', episode_step, 'step_counter, 成功结束')
# print(RL.policy_net.state_dict()['fc1.bias'])
if not learn:
print('测试结束')
else:
print('学习结束')
env.destroy()
return RL
if __name__ == '__main__':
env = Maze('train')
n_actions = env.n_actions
n_features = env.n_features
RL = DQN(n_actions,
n_features,
memory_size=256,
batch_size=8,
learning_rate=0.01,
gamma=0.9,
e_greedy=0.9,
e_greedy_increment=None,
ddqn=True,
dueling=True,
target_update_step=10,
output_graph=False)
observation = env.reset()
action = RL.choose_action(str(observation))
num_steps = 0
while True:
num_steps += 1
# fresh env
env.render()
# 下一个action
action_ = RL.choose_action(str(observation))
observation_, reward, done = env.step(action)
RL.learn(str(observation), action, reward, str(observation_),
action_)
observation = observation_
action = action_
if done:
print('Episode %d, num_steps=%d' % (episode + 1, num_steps))
break
print('game over')
env.destroy()
if __name__ == '__main__':
env = Maze()
RL = SarsaTable(actions=list(range(env.n_actions)))
env.after(100, update)
env.mainloop()
class CustomGym(Env):
"""The main OpenAI Gym class. It encapsulates an environment with
arbitrary behind-the-scenes dynamics. An environment can be
partially or fully observed.
The main API methods that users of this class need to know are:
step
reset
render
close
seed
And set the following attributes:
action_space: The Space object corresponding to valid actions
observation_space: The Space object corresponding to valid observations
reward_range: A tuple corresponding to the min and max possible rewards
Note: a default reward range set to [-inf,+inf] already exists. Set it if you want a narrower range.
The methods are accessed publicly as "step", "reset", etc...
"""
def __init__(
self,
agentXY,
goalXY,
walls=[],
pits=[],
title='Maze',
):
super(CustomGym, self).__init__()
self.env = Maze(agentXY, goalXY, walls, pits, title)
self.title = title
self.action_space = spaces.Discrete(self.env.n_actions)
self.observation_space = spaces.Box(low=0,
high=0,
shape=(4, ),
dtype=np.float32)
self.rewards = [[]]
self.variance = []
self.median = []
def step(self, action):
"""Run one timestep of the environment's dynamics. When end of
episode is reached, you are responsible for calling `reset()`
to reset this environment's state.
Accepts an action and returns a tuple (observation, reward, done, info).
Args:
action (object): an action provided by the agent
Returns:
observation (object): agent's observation of the current environment
reward (float) : amount of reward returned after previous action
done (bool): whether the episode has ended, in which case further step() calls will return undefined results
info (dict): contains auxiliary diagnostic information (helpful for debugging, and sometimes learning)
"""
s_, reward, done = self.env.step(action)
self.rewards[-1].append(reward)
if done:
self.variance.append(np.var(self.rewards[-1]))
self.median.append(np.median(self.rewards[-1]))
self.rewards.append([])
return s_, reward, done, {}
def render(self, mode='human'):
self.env.render()
def reset(self, value=1, resetAgent=True):
return self.env.reset()
def seed(self, seed=None):
"""Sets the seed for this env's random number generator(s).
Note:
Some environments use multiple pseudorandom number generators.
We want to capture all such seeds used in order to ensure that
there aren't accidental correlations between multiple generators.
Returns:
list<bigint>: Returns the list of seeds used in this env's random
number generators. The first value in the list should be the
"main" seed, or the value which a reproducer should pass to
'seed'. Often, the main seed equals the provided 'seed', but
this won't be true if seed=None, for example.
"""
np.random.seed(10)
random.seed(10)
return
def save_csv(self):
with open(f"./data/{self.title}_rewards_{time.time()}",
"w+") as my_csv:
csvWriter = csv.writer(my_csv, delimiter=',')
csvWriter.writerows(self.rewards)
with open(f"./data/{self.title}_variance_{time.time()}",
"w+") as my_csv:
csvWriter = csv.writer(my_csv, delimiter=',')
for var in self.variance:
csvWriter.writerow([var])
with open(f"./data/{self.title}_median_{time.time()}", "w+") as my_csv:
csvWriter = csv.writer(my_csv, delimiter=',')
for med in self.median:
csvWriter.writerow([med])
def destroy(self):
self.env.destroy()
while True:
# fresh env
env.render()
# RL choose action based on observation
action = RL.choose_action(str(observation))
# RL take action and get next observation and reward
observation_, reward, done = env.step(action)
# RL learn from this transition
RL.learn(str(observation), action, reward, str(observation_))
# swap observation
observation = observation_
# break while loop when end of this episode
if done:
break
# end of game
print('game over')
env.destroy()
if __name__ == "__main__":
env = Maze()
RL = QLearningTable(actions=list(range(env.n_actions)))
env.after(100, update)
env.mainloop()
#Task 3
# wall_shape=np.array([[7,4],[7,3],[6,3],[6,2],[5,2],[4,2],[3,2],[3,3],[3,4],[3,5],[3,6],[4,6],[5,6]])
# pits=np.array([[1,3],[0,5], [7,7]])
experiments = []
# # alg0 (Aynsc)
# env0 = Maze(agentXY,goalXY,wall_shape, pits)
# RL0 = rlalg0(actions=list(range(env0.n_actions)))
# data0={}
# env0.after(10, update(env0, RL0, data0, episodes))
# env0.mainloop()
# experiments = [(env0,RL0, data0)]
# alg2 (SARSA)
env2 = Maze(agentXY, goalXY, wall_shape, pits)
RL2 = rlalg2(actions=list(range(env2.n_actions)))
data2 = {}
env2.after(10, update(env2, RL2, data2, episodes))
env2.mainloop()
experiments.append((env2, RL2, data2))
# alg1 (Q-Learning)
env1 = Maze(agentXY, goalXY, wall_shape, pits)
RL1 = rlalg1(actions=list(range(env1.n_actions)))
data1 = {}
env1.after(10, update(env1, RL1, data1, episodes))
env1.mainloop()
experiments.append((env1, RL1, data1))
# alg4 (Expected Sarsa)