Pixel to policy: DQN Encoders for within & cross-game reinforcement learning