Giters

udacity / deep-reinforcement-learning

Repo for the Deep Reinforcement Learning Nanodegree program

Home Page:https://www.udacity.com/course/deep-reinforcement-learning-nanodegree--nd893

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

OUNoise should use normal distribution

opened this issue · comments

commented

The current implementation uses random.random() which I believe is uniform distribution between [0,1). This can negatively affect exploration abilities of DDPG agent, since noise will have positive bias.

commented

Let me add here that the Wikipedia page also says that the second term in the sum defining the noise, which is supposed to be dW_t=W_{t+1}-W_t must be a Wiener Process which as mentioned in property 3 needs to have Gaussian increments, hence what @lukcn has referenced above.