Shunyu Liu

I think if you set requires_grad = True, you will back propagation gradient twice, because the adv image can also be back propagation to the network which generate it.

作者的项目真的太有用了哇！ 同求论文分类筛选功能，不知道作者还更新吗？

> @liushunyu hello, windows10 or ubuntu? ubuntu

没有遇到过也，不过你这个错误看着像是安装 cloudpickle==1.1.1 就好了？ > @liushunyu 想请问下您在安装baseline的时候有遇到这种错误吗？ ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following...

@fighting-zz 不好意思阿，我没有同时安装过baseline 和gfootball，所以参考意义不是很大。我是以前用过baseline这个包，当时是按照baseline仓库的安装命令安装，后面没有用了，我记得那时候的安装是没问题的。感觉你的问题可能是baseline的一些依赖项太老，而gfootball太新了，我建议是先不管冲突以新的来，先试试代码能不能跑起来，要是代码没报错我觉得应该没问题，有报错再尝试把某些冲突的包进行降级。

@fighting-zz 嗯嗯是的

@fighting-zz 我是在没有图形化界面的unbuntu上来的，没有可视化出来看过

@fighting-zz 恭喜，是的用的虚拟环境

Ok, you answer my first question, as you said, the zero value of P@2 is because of stochasticity. but the second question？**why the P@k will decrease with increasing k？**

> Are you saying why is P@k increasing with increasing k? Not the other way round right? oh, yes