Keras implementation is not RL

[itschenxi]

It seems the Keras implementation does not use policy gradient algorithm (reinforcement learning), instead it uses supervised learning. The Pytorch implementation uses reinforcement learning. Why?