def main(argv=()):
print("hello world")
#game = make_game(int(argv[1]) if len(argv) > 1 else 0)
#humanplayer_scrolly(game)
env = Maze()
agent = GMRAgent(actions=list(range(env.n_actions)))
n_trj = 1000
reward_list = []
step_r = []
for eps in range(n_trj):
observation = env.reset()
step = 0
re_vec = []
r_episode = 0
while step < 200:
step += 1
#env.render()
#action = agent.random_action(str(observation))
state = agent.obs2state(observation)
action = agent.ActAccordingToGM(state)
observation_, reward, done, info = env.step(action)
state_ = agent.obs2state(observation_)
re_vec.append([state, action, reward, state_])
#agent.MemoryWriter(re_vec)
agent.PairWriter(state, action, reward, state_)
r_episode += reward
step_r.append(reward)
observation = observation_
if done:
print("done!")
break
#print("re_vec",re_vec)
#agent.MemoryWriter(re_vec)
reward_list.append(r_episode)
if len(step_r) > 12000:
break
# agent.plotGmemory()
# print('state',state)
# agent.MemoryReader(state)
plt.figure(1)
plt.plot(reward_list)
#plt.show()
plt.savefig("./RESULT/GQ/reward_list005.png")
reward_list = np.array(reward_list)
np.save('./RESULT/GQ/reward_list005.npy', reward_list)
temp_step_r = []
for i in range(len(step_r)):
if i < 300:
temp_step_r.append(0)
else:
temp_step_r.append(sum(step_r[(i - 290):i]) / 290)
#print("temp_step_r",temp_step_r)
plt.figure(2)
plt.plot(temp_step_r)
#plt.show()
plt.savefig("./RESULT/GQ/step_r005.png")
temp_step_r = np.array(temp_step_r)
np.save('./RESULT/GQ/temp_step_r005.npy', temp_step_r)
def main(argv=()):
print("hello world")
#game = make_game(int(argv[1]) if len(argv) > 1 else 0)
#humanplayer_scrolly(game)
env = Maze()
agent = GMRAgent(actions=list(range(env.n_actions)))
n_trj = 1000
for eps in range(n_trj):
observation = env.reset()
step = 0
re_vec = []
while step < 1000:
step += 1
env.render()
#action = agent.random_action(str(observation))
state = agent.obs2state(observation)
action = agent.ActAccordingToGM(state)
observation_, reward, done = env.step(action)
state_ = agent.obs2state(observation_)
re_vec.append([state, action, reward, state_])
observation = observation_
if done:
print("done!")
break
#print("re_vec",re_vec)
agent.MemoryWriter(re_vec)
agent.plotGmemory()
print('state', state)
agent.MemoryReader(state)
def main(argv=()):
print("hello world")
#game = make_game(int(argv[1]) if len(argv) > 1 else 0)
#humanplayer_scrolly(game)
env = Maze()
agent = GMRAgent(actions=list(range(env.n_actions)))
#n_trj = 100
n_trj = 3
reward_list = []
step_r = []
for eps in range(n_trj):
observation = env.reset()
step = 0
re_vec = []
r_episode = 0
while step < 200:
step += 1
env.render()
#action = agent.random_action(str(observation))
state = agent.obs2state(observation)
action = agent.ActAccordingToGM(state)
observation_, reward, done = env.step(action)
state_ = agent.obs2state(observation_)
re_vec.append([state, action, reward, state_])
#agent.MemoryWriter(re_vec)
agent.PairWriter(state, action, reward, state_)
r_episode += reward
step_r.append(reward)
observation = observation_
if done:
print("done!")
break
#print("re_vec",re_vec)
#agent.MemoryWriter(re_vec)
agent.plotGmemory()
plt.savefig(SAVEPATH + str(eps) + "original.png")
reward_list.append(r_episode)
t1 = 120
t2 = 100
agent.MemoryReconstruction(t1, t2)
agent.plotGmemory()
plt.savefig(SAVEPATH + str(eps) + "reconstruct.png")
break
reward_list.append(r_episode)
plt.figure(1)
plt.plot(reward_list)
#plt.show()
plt.savefig("./RESULT/Q/reward_list003.png")
reward_list = np.array(reward_list)
np.save('./RESULT/Q/reward_list003.npy', reward_list)
temp_step_r = []
for i in range(len(step_r)):
if i < 500:
temp_step_r.append(step_r[i])
else:
temp_step_r.append(sum(step_r[(i - 490):i]) / 490)
#print("temp_step_r",temp_step_r)
plt.figure(2)
plt.plot(temp_step_r)
#plt.show()
plt.savefig("./RESULT/Q/step_r003.png")
temp_step_r = np.array(temp_step_r)
np.save('./RESULT/Q/temp_step_r003.npy', temp_step_r)
if __name__ == "__main__":
# env = gym.make('MontezumaRevengeNoFrameskip-v4')
# main_MR()
env = Maze()
S_space = env.state_space()
main_MAZE(env)
def main(argv=()):
print("hello world")
#game = make_game(int(argv[1]) if len(argv) > 1 else 0)
#humanplayer_scrolly(game)
env=Maze()
agent= GMRAgent(actions=list(range(env.n_actions)))
#n_trj = 100
n_trj = 2000
reward_list=[]
step_r=[]
for eps in range(n_trj):
observation = env.reset()
step = 0
re_vec = []
r_episode =0
while step <200:
step +=1
#env.render()
#action = agent.random_action(str(observation))
state = agent.obs2state(observation)
action = agent.ActAccordingToGM(state)
observation_, reward, done, info = env.step(action)
state_ = agent.obs2state(observation_)
re_vec.append([state, action, reward, state_])
#agent.MemoryWriter(re_vec)
agent.PairWriter(state,action,reward,state_)
r_episode += reward
step_r.append(reward)
observation = observation_
if done:
print("done!")
break
#print("re_vec",re_vec)
#agent.MemoryWriter(re_vec)
# agent.plotGmemory()
# plt.savefig(SAVEPATH+str(eps)+"original_substract.png")
reward_list.append(r_episode)
t1=120
t2=100
agent.MemoryReconstruction(t1,t2)
# plt.figure(0)
# agent.plotGmemory()
# plt.savefig(SAVEPATH+str(eps)+"reconstruct_substract.png", dpi=1080)
# plt.close(0)
# plt.figure(3)
# agent.plotAbastract()
# plt.savefig(SAVEPATH+str(eps)+"abstract.png", dpi=1080)
# plt.close(3)
# plt.figure(4)
# agent.plotReconPath()
# plt.savefig(SAVEPATH+str(eps)+"reconpath.png", dpi=1080)
# plt.close(4)
if len(step_r)>12000:
break
# agent.plotGmemory()
plt.figure(1)
plt.plot(reward_list)
#plt.show()
plt.savefig(SAVEPATH+"reward_list005.png")
reward_list=np.array(reward_list)
np.save(SAVEPATH+'reward_list005.npy',reward_list)
temp_step_r=[]
for i in range(len(step_r)):
if i<200 :
if step_r[i]>100:# 这个纯粹为了画出来的曲线在同一个尺度上，便于和其它算法比较，没有其它含义
temp_step_r.append(0)
else:
temp_step_r.append(step_r[i])
else:
temp_step_r.append(sum(step_r[(i-190):i])/190)
#print("temp_step_r",temp_step_r)
plt.figure(2)
plt.plot(temp_step_r)
#plt.show()
plt.savefig(SAVEPATH+"step_r005.png")
temp_step_r=np.array(temp_step_r)
np.save(SAVEPATH+'temp_step_r005.npy',temp_step_r)