Bipedal Walker env from OpenAI enviroment. Midterm assignment for Move37 from theshool.ai

This is a state of the art algorithm that helps us find the optimal policy way more faster that other algorithms e.g. Q-learning This algorithm is explain in detail by colin skow ands is part of Move37, a theschool.ai course. I try the algorithm and it starts to converge at the 300 step aprox., but i wanted to be able to save the values of the weight matrix, so since the ars.py is using Numpy to create the matrix, we used a built-in function to save it into a csv so we can use this values later to achive similar results.