windowsでTensorFlow その7

2546 ワード
強化学習 TensorFlow1.0 OpenAIGym 強化学習 テキストリンク

概要

windowsのTensorFlowの環境で、OpenAiやってみた。
手でやってみた。

写真

環境

windows 7 sp1 64bit
anaconda3
tensorflow 1.0
pyqt5

インストール

pip install gym

サンプルコード

from __future__ import print_function
import sys, gym

env = gym.make('Acrobot-v1' if len(sys.argv) < 2 else sys.argv[1])
if not hasattr(env.action_space, 'n'):
    raise Exception('Keyboard agent only supports discrete action spaces')
ACTIONS = env.action_space.n
ROLLOUT_TIME = 1000
SKIP_CONTROL = 0
human_agent_action = 0
human_wants_restart = False
human_sets_pause = False

def key_press(key, mod):
    global human_agent_action, human_wants_restart, human_sets_pause
    if key == 0xff0d:
        human_wants_restart = True
    if key == 32:
        human_sets_pause = not human_sets_pause
    a = int(key - ord('0'))
    if a <= 0 or a >= ACTIONS:
        return
    human_agent_action = a

def key_release(key, mod):
    global human_agent_action
    a = int(key - ord('0'))
    if a <= 0 or a >= ACTIONS:
        return
    if human_agent_action == a:
        human_agent_action = 0
env.render()
env.unwrapped.viewer.window.on_key_press = key_press
env.unwrapped.viewer.window.on_key_release = key_release

def rollout(env):
    global human_agent_action, human_wants_restart, human_sets_pause
    human_wants_restart = False
    obser = env.reset()
    skip = 0
    for t in range(ROLLOUT_TIME):
        if not skip:
            a = human_agent_action
            skip = SKIP_CONTROL
        else:
            skip -= 1
        obser, r, done, info = env.step(a)
        env.render()
        if done:
            break
        if human_wants_restart:
            break
        while human_sets_pause:
            env.render()
            import time
            time.sleep(0.1)
print ("ACTIONS={}".format(ACTIONS))
print ("Press keys 1 2 3 ... to take actions 1 2 3 ...")
print ("No keys pressed is taking action 0")

while 1:
    rollout(env)