23a204d75f91c85bb542269447dbfa2164c695ce,ch13/wob_click_train.py,,,#,27

Before Change

    env = wob_vnc.MiniWoBCropper(env)

    env.configure(remotes=REMOTE_ADDR)
    obs = env.reset()

    net = model_vnc.Model(input_shape=(3, wob_vnc.HEIGHT, wob_vnc.WIDTH),
                          n_actions=env.action_space.n)
    print(net)

After Change

    agent = ptan.agent.PolicyAgent(lambda x: net(x)[0], cuda=args.cuda,
                                   apply_softmax=True)
    exp_source = ptan.experience.ExperienceSourceFirstLast(
        [env], agent, gamma=GAMMA, steps_count=REWARD_STEPS, vectorized=True)

    // obs, reward, done, info = step_env(env, env.action_space.sample())
    // obs_v = Variable(torch.from_numpy(np.array(obs)))
    // r = net(obs_v)
    // print(r[0].size(), r[1].size())

    for idx, exp in enumerate(exp_source):
        print(exp)
        if idx > 100:
            break
        time.sleep(0.5)

    pass

Instances