def eval_q_net():
"""
连跑训练地图各5000次
日志输出单步成功率，回合成功率
:return:
"""
brain = BrainDQN()
brain.epsilon = 0
for index in range(len(_MAP_LIST)):
maze = env.Maze(_MAP_LIST,
is_show=True,
is_loop=False,
map_index=index)
maze.effective_epsilon = 1
for i in range(1001):
observation, key_observation = maze.reset()
while True:
maze.render(False)
action, is_random = brain.getAction(observation,
key_observation)
next_observation, next_key_observation, r, done = maze.step(
action, is_random)
observation = next_observation.copy()
key_observation = next_key_observation.copy()
if done:
break
def main(is_debug):
brain = BrainDQN()
if is_debug:
brain.epsilon = 0
maze = env.Maze(_MAP_LIST, is_show=True)
while True:
observation, key_observation = maze.reset()
while True:
maze.render(is_debug)
action, is_random = brain.getAction(observation, key_observation)
next_observation, next_key_observation, r, done = maze.step(
action, is_random)
brain.setPerception(observation, key_observation, next_observation,
next_key_observation, action, r, done)
observation = next_observation.copy()
key_observation = next_key_observation.copy()
if done:
break
def eval_q_net():
maze = env.Maze(_MAP1, is_show=True)
memory = ReplayMemory()
memory.load("eval_data")
brain = BrainDQN()
# 评估时不随机
brain.epsilon = 0
r_list = []
for data in memory.memory:
s, action, r, s_, done = data[0], data[1], data[2], data[3], data[4]
action, _ = brain.getAction(s)
maze.set_observation(s)
_, r_eval, done_eval = maze.step(action, False)
r_list.append(r_eval - r)
logging.info(sum(r_list) / len(r_list))
plt.plot(np.arange(len(r_list)), r_list)
plt.ylabel('reward loss')
plt.xlabel('sample')
plt.show()
def main():
brain = BrainDQN()
maze = env.Maze(_MAP1, is_show=True)
while True:
observation = maze.reset()
while True:
maze.render()
action, is_random = brain.getAction(observation)
next_observation, r, done = maze.step(action, is_random)
brain.setPerception(observation, next_observation, action, r, done)
observation = next_observation.copy()
if done:
break
def gen_eval_data_manually():
maze = env.Maze(_MAP1, is_show=True)
key_map = {
"1": env.LEFT_DOWN,
"2": env.DOWN,
"3": env.RIGHT_DOWN,
"4": env.LEFT,
"5": env.STOP,
"6": env.RIGHT,
"7": env.LEFT_UP,
"8": env.UP,
"9": env.RIGHT_UP
}
memory = ReplayMemory()
memory.load("eval_data")
done = True
while True:
if done:
observation = maze.reset()
s = input('input your path. Enter "exit" to close\n')
if s == "exit":
break
next_observation, r, done = maze.step(key_map[s], True)
memory.push_back(
(observation.copy(), key_map[s], r, next_observation.copy(), done))
observation = next_observation.copy()
maze.render(is_sleep=True)
memory.dump("eval_data")