This is implementation of a RL agent that learns to play the Banana Collector game.

The environment is a plane with yellow and blue bananas in the plane. The bananas are kept randomly in the plane.

The goal is to collect as many yellow bananas as possible while avoiding blue bananas.

The state space has 37 dimensions and contains the agent's velocity, along with ray-based perception of objects around agent's forward direction. Given this information, the agent has to learn how to best select actions.

There are 4 possible actions:

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

The reward function looks something like below -

1. +1 for interaction with yellow banana
2. -1 for interaction with blue banana

After creating the conda environment using the commands mentioned below and activating it, you will be able to run the code in this repository.

conda create --name bananas python=3.6

conda activate bananas

Note - If you are having a problem with loading unityagents library in python or problem with loading the Banana environment, the following these steps of creating the environment, activating the created environment and installing the requirements mentioned in requirements.txt may help. Sometimes, the environment may not load from jupyter notebook, so run it from python script instead.

pip install -r requirements.txt

Download the environment from one of the links below. You need only select the environment that matches your operating system:

• Linux: click here

• Mac OSX: [click here](https://s3-us-west-1.amazonaws.com/udacity-drlnd/P1/Banana/Banana.app.zip

• Windows (32-bit): click here

• Windows (64-bit): click here

• (For Windows users) Check out this link if you need help with determining if your computer is running a 32-bit version or 64-bit version of the Windows operating system.

• (For AWS) If you'd like to train the agent on AWS (and have not enabled a virtual screen), then please use this link to obtain the environment.

Simply run the train.py script using the following command

python train.py

The above script runs the training for 1800 episodes and saves the data in the form of a dictionary as a file named as "checkpoint.pth"

Running the test.py runs loads the model saved by the train.py script and runs 10 episodes with the trained Q Network which takes a state and returns best action that maximises the action value function.

python test.py