INDEX

• Project Purpose
• Environment
  • How to set env
• Implements
  • Preprocessing
  • Algorithm
  • Model
• Performance

Project Purpose

The ultimate goal of this project is to implement atari game player by reinforcement learning algorithm DQN.

Environment

• OS : Window 11
• Python : 3.6
• Use Anaconda3
• Use Pytorch
• Use gym(Open AI)

How to set env

1. You should install Anaconda3 -> Anaconda download link
2. Create Anaconda virtual env and turn on the env

# create env
conda create -n atari_openai python=3.6
# activate env
conda activate atari_test

3. Install required pakages by conda

# pytorch(cpu or gpu)
conda install pytorch torchvision torchaudio cpuonly -c pytorch # if your pc use only cpu
conda install pytorch torchvision torchaudio cudatoolkit=11.1 -c pytorch -c conda-forge # gpu

# opencv 
conda install -c conda-forge opencv

# gym
conda install -c conda-forge gym

# atari_py
conda install -c conda-forge atari_py

4. Install Roms(Roms download Link) and unzip a file
5. Enter the following command.

python -m atari_py.import_roms <path> # path ex) ~\Roms\ROMS

Implements

*I referred to the paper(Playing Atari with Deep Reinforcement Learning - DeepMind).

Preprocessing

Raw Atari frame ,which are 210 X 160 pixel images with a 128 color palette, is computationally demanding. So we should apply preprocessing for reducing the input dimensionality. The raw frames are preprocessed by converting RGB to gray-scale and resizing it to a 84 x 84 pixel image using opencv library.

It is implemented in utils.py as make_env function.

Algorithm

It is implemented in utils.py as experience function.

Model

It is implemented in network.py.

Performance

• Result of learning 7 atari games as DQN
• Hyper parameter epsilon tuning for BreakOut game