The project is explained here and describes how to do Reinforcement Learning with q-learning and neural networks.
- qlearning_FrozenLake.ipynb notebook explore q-learning technique in FrozenLake openai gym environment:
- A neural network was also used to teach an agent to land on a specific position. The following agent was trained for 3k episodes:
- The same architecture was used in the CartPole openai gym environment where a pole is attached by an un-actuated joint to a cart, which moves along a frictionless track:
The architecture and methods of the agent have been designed to adapt to other open gym environments.
The project is explained here
It explains all the steps involved in building a neural network to play Atari games; how to: get the OpenAi Gym environment, preprocess the environment to feed the CNN, implement the CNN, create the training loop.