training.py 文件源码

def test(args, env, agent):
    if args.record:
        if 'env' in vars(args):
            env = wrappers.Monitor(env, './videos/' + args.env + str(time()) + '/')
        else:
            env = wrappers.Monitor(env, './videos/' + str(time()) + '/')
    test_rewards = []
    test_start = time()
    test_steps = 0
    for iteration in range(1, 1 + args.n_test_iter):
        state = env.reset()
        iter_rewards = 0.0
        done = False
        while not done:
            test_steps += 1
            action, _ = agent.forward(state)
            state, reward, done, _ = env.step(action)
            iter_rewards += reward
        test_rewards.append(iter_rewards)
    print_stats('Test', test_rewards, args.n_test_iter,
                time() - test_start, test_steps, 0, agent)
    return test_rewards