Reinforcement Learning, tensorflow custom python environment

Question

I was trying to implement a custom python environment for tensorflow. So my _step method while returning ts.transition(np.array(observation = self._state, reward=reward, d