py_environment'time_step'与'time_step_spec'不匹配

时间:2019-07-29 18:24:30

标签: tensorflow-agents

我已经通过tf代理创建了一个自定义的pyenvironment。但是我无法使用py_policy.action验证环境或在其中采取措施 我对time_step_specs之外的内容感到困惑

我曾尝试通过tf_py_environment.TFPyEnvironment转换为tf_py_environment,并成功地通过tf_policy采取了措施,但是我仍然对区别感到困惑。

import abc
import numpy as np
from tf_agents.environments import py_environment
from tf_agents.environments import tf_environment
from tf_agents.environments import tf_py_environment
from tf_agents.environments import utils
from tf_agents.specs import array_spec
from tf_agents.environments import wrappers
from tf_agents.trajectories import time_step as ts
from tf_agents.policies import random_tf_policy
import tensorflow as tf
import tf_agents

class TicTacToe(py_environment.PyEnvironment):
   def __init__(self,n):
    super(TicTacToe,self).__init__()
    self.n = n
    self.winner = None
    self._episode_ended = False
    self.inital_state = np.zeros((n,n))
    self._state = self.inital_state
    self._observation_spec = array_spec.BoundedArraySpec(
        shape = (n,n),dtype='int32',minimum = -1,maximum = 1,name = 
'TicTacToe board state spec')
    self._action_spec = array_spec.BoundedArraySpec(
        shape = (),dtype = 'int32', minimum = 0,maximum = 8, name = 
'TicTacToe action spec')

def observation_spec(self):
    return self._observation_spec

def action_spec(self):
    return self._action_spec

def _reset(self):
    return ts.restart(self.inital_state)

def check_game_over(self):
    for i in range(self.n):
        if (sum(self._state[i,:])==self.n) or 
(sum(self._state[:,i])==self.n):
            self.winner = 1
            return True

        elif (sum(self._state[i,:])==-self.n) or 
    (sum(self._state[:,i])==-self.n):
            self.winner = -1
            return True

    if (self._state.trace()==self.n) or 
(self._state[::-1].trace()==self.n):
        self.winner = 1
        return True
    elif (self._state.trace()==-self.n) or (self._state[::-1].trace()==- 
   self.n):
        self.winner = -1
        return True

    if not (0 in self._state):
        return True

def _step(self,action):
    self._state[action//3,action%3]=1
    self._episode_ended = self.check_game_over

    if self._episode_ended==True:
        if self.winner == 1:
            reward = 1
        elif self.winner == None:
            reward = 0
        else:
            reward = -1
        return ts.termination(self._state,dtype = 'int32',reward=reward)
    else:
        return ts.transition(self._state,dtype = 'int32',reward = 
0.0,discount = 0.9)

env = TicTacToe(3)
utils.validate_py_environment(env, episodes=5)

这是我得到的错误:

ValueError跟踪(最近一次通话最近)  在 ----> 1 utils.validate_py_environment(env,episodes = 5)

validate_py_environment中的C:\ Users \ bzhang \ AppData \ Local \ Continuum \ anaconda3 \ lib \ site-packages \ tf_agents \ environments \ utils.py(环境,情节)      58提高ValueError(      59'给出time_step:%r与预期的time_step_spec不匹配:%r'% ---> 60(time_step,time_step_spec))      61      62行动= random_policy.action(time_step).action

ValueError:给定time_step:TimeStep(step_type = array(0),reward = array(0。,dtype = float32),Discount = array(1。,dtype = float32),observation = array([ [0.,0.,0.],        [0.,0.,0.],        [0.,0.,0.]]))与预期的time_step_spec不符:TimeStep(step_type = ArraySpec(shape =(),dtype = dtype('int32'),name ='step_type'), reward = ArraySpec(shape =(),dtype = dtype('float32'),name ='reward'),discount = BoundedArraySpec(shape =(),dtype = dtype('float32'),name ='discount',最小值= 0.0,最大值= 1.0),观察值= BoundedArraySpec(shape =(3,3),dtype = dtype('int32'),name ='TicTacToe板状态规范',最小值= -1,最大值= 1))

1 个答案:

答案 0 :(得分:0)

您的观察结果不符合规范,您需要将dtype=np.int32传递给np数组以确保类型匹配。