Q-learning with neural network

I am trying to implement Q-learning together with a neural network (known as deep Q-learning).

I know the theory but cant understand how to implement it in Matlab.

Can anyone assist with an Example code on this?

Thanks in advance!

Answer 1

[Agnish Dutta]

MATLAB's reinforcement learning toolbox has tools for implementing a variety of RL algorithms such as Deep Q-Network (DQN), Advantage Actor Critic (A2C), Deep Deterministic Policy Gradients (DDPG), and other built-in algorithms.

1) Consider going through the following tutorial to get an idea about running a simple Q learning Agent in an MDP environment. It explains the process of creating an MDP, and agent and then training the agent on the MDP environment.

https://www.mathworks.com/help/reinforcement-learning/ug/train-reinforcement-learning-agent-in-mdp-environment.html

2) The above is geared towards Q learning. To create and train a DQN agent, kindly go through the following:

Create a DQN learning agent - https://www.mathworks.com/help/reinforcement-learning/ref/rldqnagent.html

You'll need to set the critic network and agent options parameters correctly. The following documents show how that can be done:

• Setting the Critic network - https://www.mathworks.com/help/reinforcement-learning/ref/rlrepresentation.html

• Setting the agent options - https://www.mathworks.com/help/reinforcement-learning/ref/rldqnagentoptions.html

3) The next thing to do is to train the agent given the environment. This can be done as depicted in the following document. It contains details about the different training options, you can use to achieve desired performance:

https://www.mathworks.com/help/reinforcement-learning/ref/rltrainingoptions.html