Atari

Research Playground built on top of OpenAI's Atari Gym, prepared for implementing various Reinforcement Learning algorithms.

It can emulate any of the following games:

['Asterix', 'Asteroids',

'MsPacman', 'Kaboom', 'BankHeist', 'Kangaroo',

'Skiing', 'FishingDerby', 'Krull', 'Berzerk',

'Tutankham', 'Zaxxon', 'Venture', 'Riverraid',

'Centipede', 'Adventure', 'BeamRider', 'CrazyClimber',

'TimePilot', 'Carnival', 'Tennis', 'Seaquest',

'Bowling', 'SpaceInvaders', 'Freeway', 'YarsRevenge',

'RoadRunner', 'JourneyEscape', 'WizardOfWor',

'Gopher', 'Breakout', 'StarGunner', 'Atlantis',

'DoubleDunk', 'Hero', 'BattleZone', 'Solaris',

'UpNDown', 'Frostbite', 'KungFuMaster', 'Pooyan',

'Pitfall', 'MontezumaRevenge', 'PrivateEye',

'AirRaid', 'Amidar', 'Robotank', 'DemonAttack',

'Defender', 'NameThisGame', 'Phoenix', 'Gravitar',

'ElevatorAction', 'Pong', 'VideoPinball', 'IceHockey',

'Boxing', 'Assault', 'Alien', 'Qbert', 'Enduro',

'ChopperCommand', 'Jamesbond']

Check out corresponding Medium article: Atari - Reinforcement Learning in depth 🤖 (Part 1: DDQN)

Purpose

The ultimate goal of this project is to implement and compare various RL approaches with atari games as a common denominator.

Usage

Clone the repo.
Go to the project's root folder.
Install required packagespip install -r requirements.txt.
Launch atari. I recommend starting with help command to see all available modes python atari.py --help.

DDQN

Hyperparameters

* GAMMA = 0.99 * MEMORY_SIZE = 900000 * BATCH_SIZE = 32 * TRAINING_FREQUENCY = 4 * TARGETNETWORKUPDATE_FREQUENCY = 40000 * MODELPERSISTENCEUPDATE_FREQUENCY = 10000 * REPLAYSTARTSIZE = 50000 * EXPLORATION_MAX = 1.0 * EXPLORATION_MIN = 0.1 * EXPLORATION_TEST = 0.02 * EXPLORATION_STEPS = 850000

Model Architecture

Deep Convolutional Neural Network by DeepMind

* Conv2D (None, 32, 20, 20) * Conv2D (None, 64, 9, 9) * Conv2D (None, 64, 7, 7) * Flatten (None, 3136) * Dense (None, 512) * Dense (None, 4) Trainable params: 1,686,180