Proximal Policy Optimization - Week 5

This week, I focus on how to store the transition into experience replay buffer. The previous codebase only supports the discrete task, so I spent some time in the continuous task.

With the help of zoq, I got two solutions on how to store the continuous task's action. One is using the std::vector to store the action, while the other is using compact the attribute of action into one column. I tried both of them, I choose the std::vector in the end. There are some misunderstanding during the coding period, it is due to my careless.

Thanks for reading :). What's more, I updated my travel note, there are many beautiful sceneries. If you are interested in it, please refer to the page.