Deep-Reinforcement-Learning-with-Double-Q-Networks-on-Atari-Games

This project uses deep RL to train an agent that can play Atari game named Space Invaders. We use openai gym for environment and keras-rl2 libraries for the agent

Deep Reinforcement Learning

With deep reinforcement learning, we use deep neural networks to estimate the q-values that are used to take an action (used in policy). In many practical decision making problems, the states s of the MDP are high-dimensional (eg. images from a camera or the raw sensor stream from a robot) and cannot be solved by traditional RL algorithms. Deep reinforcement learning algorithms incorporate deep learning to solve such Maps a, often representing the policy π ( a | s ) or other learned functions as a neural network, and developing specialized algorithms that perform well in this setting.

Installations

The following steps can be considered before installing gym and atari-py in Windows Anaconda. [Linux Installation is straight forward]

Step 1: Create a new environment in anaconda:

conda create -n <env_name> python=3.9
conda activate <env_name?

Step 2: [Depends on Visual Studio]. If Visual Studio is not present please download and install

• Download VS build tools here
• Run the VS build setup and select "C++ build tools" and install it.

Step 3: Packages Installation in the created environment [Using Pip]

pip install tensorflow
pip install cmake
pip install atari-py
pip install gym
pip install gym[atari]
pip install keras-rl2

Step 4: With the latest atari-py verions, only Tetris game is available. To get all the games:

• Download the ROMS from this link
• Unrar the folder at any location
• Run the below code in the conda prompt

python -m atari_py.import_roms <path to folder where ROMS are unrared>

Step 5: Enjoy Coding !!!