Naive implementation of Monte-Carlo Policy-Gradient Control. CartPole-v0 has been used here as the environment.
The algorithm is given below.
There is one trick though. The return, G, is normalized. This helps the algorithm to have numerical stability.