jwk1rose / rl_learning Goto Github PK

if (epoch+1)%50==0: # 测试
    n_actions=5
    action_img = ['↑', '→', '↓', '←','o'] # 五个方向对应的图标
    print('--- test epoch ---')
    # 固定策略(对于每个state固定)
    for x in range(5):
        for y in range(5):
            s = torch.tensor((x, y), dtype=torch.float)
            q_value = torch.zeros((n_actions), dtype=torch.float) # 分配五个活动对应的空间，下面分别计算
            for act in range(n_actions):
                nxt_s_in = torch.cat((s, torch.tensor(act).unsqueeze(0)))
                # print(nxt_s_in)
                q_value[act] = q_net(nxt_s_in) 
            max_idx = torch.argmax(q_value)
            print(action_img[max_idx], end=' ')
        print()