README.md

Reinforcement Learning: Deeq Q Learning Network (DQN) Agent playing Atari Breakout

• Training a vision-based agent with the Deep Q Learning Network (DQN) in Atari's Breakout environment, implementation in Tensorflow.

Environment

• < Python 3.7 >
• < OpenAI Gym >
  • Install the OpenAI Gym Atari environment: $ pip3 install opencv-python gym "gym[atari]"
  • Atari environment used: BreakoutNoFrameskip-v4
• < Tensorflow r.1.12.0 >

Implementation

• Deep Q Learning Network with the following improvements:
  • Experience Replay
  • Fixed Target Q-Network
  • TD error loss function with: Q_target = reward + (1-terminal) * (gamma * Q_max(s’)
)
• DQN network Settings (in agent_dqn.py):

File Description

.
├── ./
|   ├── agent_dqn.py ─────────── DQN model
|   ├── atari_wrapper.py ─────── Atari wrapper
|   ├── environment.py ───────── Gym wrapper
|   ├── runner.py ────────────── Main program for training and testing
|   └── Readme.md ────────────── This file
└── model/
	├── dqn_learning_curve_compare.png ──────── Figure 1  
	├── dqn_best_setting.png ────────────────── Figure 2
	├── dqn_learning_curve.png ──────────────── Figure 3
	├── checkpoint ──────────────────────────── Tensorflow model check point
	├── model_dqn-25581.data-00000-of-00001 ─── Tensorflow model data
	├── model_dqn-25581.meta ────────────────── Tensorflow model meta
	└── model_dqn-25581.index ───────────────── Tensorflow model index

Usage

• Traing the DQN Agent: $ python3 runner.py --train_dqn
• Testing the DQN Agent: $ python3 runner.py --test_dqn
• Testing the DQN Agent with gameplay rendering: $ python3 runner.py --test_dqn --do_render

Learning Curve

• Single learning curve:
• With different plotting window: