Discussing the article: "MQL5 Wizard Techniques you should know (Part 62): Using Patterns of ADX and CCI with Reinforcement-Learning TRPO"
You are missing trading opportunities:
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
Registration
Log in
You agree to website policy and terms of use
If you do not have an account, please register
Check out the new article: MQL5 Wizard Techniques you should know (Part 62): Using Patterns of ADX and CCI with Reinforcement-Learning TRPO.
The ADX Oscillator and CCI oscillator are trend following and momentum indicators that can be paired when developing an Expert Advisor. We continue where we left off in the last article by examining how in-use training, and updating of our developed model, can be made thanks to reinforcement-learning. We are using an algorithm we are yet to cover in these series, known as Trusted Region Policy Optimization. And, as always, Expert Advisor assembly by the MQL5 Wizard allows us to set up our model(s) for testing much quicker and also in a way where it can be distributed and tested with different signal types.
It used to be, that in order to prompt any computer program for a usable or practical response, this response, had to be hand-coded into the program. Essentially, the if-clause was core to programming most programs. And if you think about it, the dependency on if-clauses meant that the user input data or data being processed by the program had to be in certain categories. It had to be discrete. Therefore, it can be argued, for the most part, that our development and use of discrete data was in response to programming constraints and not pertinent to the data or problem being solved.
And then came OpenAI in the fall of 2023 with their first public GPT, and all this changed. The development of transformer networks and GPTs did not happen overnight, as the first perceptrons were developed in the late 60s, but it is safe to say the launch of ChatGPT was a significant milestone. With the wide adoption of large-language-models, it has become abundantly clear tokenization, word-embedding, and of course self-attention are critical components in allowing models to scale with what they can process. No more if-clauses. It is with this backdrop of using tokenization and word-embedding in making network inputs as continuous as possible that we also made the inputs of our supervised-learning MLP ‘more continuous’.
Author: Stephen Njuki