A solution to the first project of Udacities Reinforcement Learning course.

This projects runs from Jupyter and has all of the code required in order to train and play back an agent.

The goal of the project is to train an agent to navigate and pick up as many yellow bannanas as possible.

You will receive a reward of +1 for collecting a yellow bannana and a reward of -1 for collecting a blue bannana.

The state space consists of 37 dimentions and has a ray-based perception of objects in the agents forward direction.

There are 4 different actions the agent can use which relates to:

• 0 Forward
• 1 Backward
• 2 Left
• 3 Right

This project is solved if the agent scores more than 13 points over 100 consecutive episodes.

You will need to install Anaconda in order to run through the following dependency instructions.

All of the dependencies for this project can be found at: https://github.com/udacity/deep-reinforcement-learning#dependencies

Inside of the Jupyter notebook, run the first code block to import all the necessary libraries to run the rest of the code blocks.

The second code block loads the Unity environement and specifies the brain name.

Deep_Q_Network (the third code block) is the deep network that is used within this project. It consists of dynamic hidden layers (so that you can change it while finding the best combination easier) and Relu activations functions.

Fourth code block is the Agent. The agent is used to determine the actions to take as well as interacts with the deep network to learn and better determine actions to increase environment rewards.

The training block runs through the specified episodes, receives the actions from the agent and passes it back to the environment. It also saves the score and model if the score is more than what is specified in the checkpoint.