I have a basic implementation of actor-critic based on keras. It works for standards open ai gym.
def __init__(self, alpha, beta, gamma=0.99, n_actions=4,