Discussing the article: "Neural networks are easy (Part 59): Dichotomy of Control (DoC)"
think about it...!
how many PHDs are working at goldmansachs? or hfts, or quantfund firms, !
if only it was THIS easy !!!
I wanted to say thanks to the author for the huge amount of ideas, it's a Klondike for experimentation.
Also I think that the articles are suitable as examples of possible methods for training neural networks, but not for practice. I really appreciate the work invested in the author's own library for creating and training neural networks and even with the use of video cards, but it can not be used in any way for practical purposes, and even less to compete with tensorflow, keras, pytorch - Actually all models trained with these libraries can be used directly in mql5 using the onnx format.
I will gradually apply the author's ideas with the help of these modern libraries.
Also it is necessary to select indicators for input data for training neural networks, I have the most successful is bollinger bands, and I use 48 such indicators as input data with different settings for recurrent networks like LSTM. But this is not a guarantee of success, I also train 28 currency pairs at a time and choose the best ones, but this is not a guarantee of success. Then you need to run at least 20 times the training procedure, changing the number of layers and their settings in neural networks, and at each stage select the best models that have shown themselves well in the strategy tester, and remove the worst, and only then you can achieve reasonable results in practice.
At the end we just choose for example the best 9 pairs out of 28 and trade them on a real account, at the same time the Expert Advisor should also have in its arsenal mani-management, it will not hurt the grid also, that is, we use neural networks as assistants to good ideas of advisors without neural networks, thus making them smart already.
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
Check out the new article: Neural networks are easy (Part 59): Dichotomy of Control (DoC).
In the previous article, we got acquainted with the Decision Transformer. But the complex stochastic environment of the foreign exchange market did not allow us to fully implement the potential of the presented method. In this article, I will introduce an algorithm that is aimed at improving the performance of algorithms in stochastic environments.
The dichotomy of control is the logical basis of Stoicism. It implies an understanding that everything that exists around us can be divided into two parts. The first one is subject to us and is completely under our control. We have no control over the second one and events will happen regardless of our actions.
We are working with the first area, while taking the second one for granted.
The authors of the "Dichotomy of Control" method implemented similar postulates into their algorithm. DoC allows us to separate what is under the control of strategy (action policy) and what is beyond its control (environmental stochasticity).Author: Dmitriy Gizlyk