2048-reinforcement-learning

The repository that hosts the code that teaches a reinforcement learning - based bot (based on policy gradients method) to play 2048 The state of the board is represented as an image, and convolutional neural network acts as a reinforcement learning agent. There are 4 actions that agent can choose (up, down, left, right). Main training loop runs via agent.py, which visualizes the board and trains the agent