Discussing the article: "MQL5 Wizard Techniques you should know (Part 36): Q-Learning with Markov Chains"

[MetaQuotes]

Check out the new article: MQL5 Wizard Techniques you should know (Part 36): Q-Learning with Markov Chains.

Reinforcement Learning is one of the three main tenets in machine learning, alongside supervised learning and unsupervised learning. It is therefore concerned with optimal control, or learning the best long-term policy that will best suit the objective function. It is with this back-drop, that we explore its possible role in informing the learning-process to an MLP of a wizard assembled Expert Advisor.

Custom signal classes for wizard assembled Expert Advisors can take on various roles, that are worth exploring, and we continue this quest by examining how the Q-Learning algorithm when paired with Markov Chains can help refine the learning process of a multi-layer-perceptron network. Q-Learning is one of the several (approximately 12) algorithms of reinforcement-learning, so essentially this is also a look at how this subject can be implemented as a custom signal and tested within a wizard assembled Expert Advisor.

So, the structure for this article will flow from what is reinforcement-learning, dwell on the Q-Learning algorithm and its cycle stages, look at how Markov chains can be integrated into Q-Learning, and then conclude as always with strategy tester reports. Reinforcement learning can be utilized as an independent signal generator because its cycles (‘episodes’) are in essence a form of learning that quantifies results as ‘rewards’ for each of the ‘environments’ the ‘actor’ is involved in. These terms in quotes are introduced below. We, however, are not using reinforcement learning as a raw signal, but rather are relying on its abilities to further the learning process by having it supplement a multi-layer-perceptron.

Author: Stephen Njuki