Supervisor: Dr. Penny Kyburz

This project allows users to use image-state in a specific game: Ice Hockey (Atari).

This project also combined the YOLO to extract the location as the state in a specific game: Ice Hockey (Atari).

Algorithms: GAIL, off-policy PPO

Youtube:https://www.youtube.com/watch?v=rgYOhhg0DUk

Warnning

• This project only supports Mac user and Linux user
• This project uses Python3

Step 0

Install Anaconda

Website (Follow the instruction from this link)

Please ensure you have installed the anaconda environment.

Install Pycharm

Website (Follow the instruction from this link) Please ensure you have installed the Pycharm environment

Setup the insterpreter in Pycharm

Open our code with Pycharm

1. Click "Pycharm" (Top menu)
2. Click "Preference"
3. Find "Project Interpreter" under your project
4. Add "Conda Envionment" one by choosig the "Existing environment"
5. Add address like "/User/USERNAME/anaconda3/bin/python"
6. Finally click "Apply"

Step 1

Edit file before next step

1. Find file and open the file "DropTheGame/Demo/StateClassifier/darknet.py"

2. Go to line 48.

3. If you are using Linux:

   lib = CDLL("StateClassifier/libdarknet_linux.so",RTLD_GLOBAL)
   

   If you are using Mac:

   lib = CDLL("StateClassifier/libdarknet_mac.so",RTLD_GLOBAL)
   

pip

pip install -r /DropTheGame/Demo/requirement.txt

Test our code on Pycharm

You can run "Start.py" directly. We have added some example codes under the main function. Uncomments any of them to try.

Test our code on Terminal

Firstly, you have to make sure you are inside "DropTheGame/Demo"

python Start

Test our code on Python Console

The functions listed in Start.py can be called as following

from Start import IceHockey
IH = IceHockey()
IH.AIplay(True,"loc")

conda

conda install -c akode gym
conda install -c conda-forge opencv

brew

brew install swig

Check Gym

from gym import envs
gym.envs.registry

Play Gym

Gym play function

If you used Spyder

Restar Spyder every time once a new tool is installed by “pip install”

Issues

print(deEnv.unwrapped.actions_space) #useless

Method 2:install via homebrew https://www.pyimagesearch.com/2016/12/19/install-opencv-3-on-macos-with-homebrew-the-easy-way/

Acknowledgments

The GAIL and PPO codes are modified based on part of OpenAI Baseline