Machine learning in trading: theory, models, practice and algo-trading - page 228
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
In short, -
1. Create an algorithm that collects streams of values of any parameters (data) we need and runs them through the ring buffer.
2. We pass the streams of values stored in the ring buffer through a special filter, which generalizes them to the ranges of these values.
3. A generalized (with the help of ranges) digital model of the nature of each parameter's value change in the ring buffer is created, and written in the appropriate format.
4. This model is sent to the statistical algorithm that collects these models.
5. We loop through the database containing models (signatures) of how our parameters values change, and find the model that best fits the current situation.
6. A decision is made about the behavior of the system in the situation captured in this signature (model).
I will formulate it more precisely later.
7. Testing, and loss of the deposit. Because it is impossible to simply gather any streams of values and make an Expert Advisor on them. These streams need to be analyzed, you need to write your own bikes to determine if each of them or some combination of them can be trusted.
For example, you can take two MovingAverage indicators and go through steps 2-6 with them without problems. The error will appear only on the seventh step.
So there is another step somewhere between 1 and 6 (everyone decides where and how to do it), which discards unsuitable data streams. A huge part of this thread on the forum is devoted to analyzing different ways to do this.
An approximation is a generalization of values. That is, is the conclusion of different data values inside a selected range? Further, you can create a numerical model that summarizes the change in a value over a period of time. By collecting these models, you can create statistics on which to base your decisions and choices of actions.
Am I going in the right direction?
Briefly.
1. We create an algorithm, which collects streams of values of any parameters (data) we need and runs them through a ring buffer.
2. Passing flows of values stored in the ring buffer through a special filter, which summarizes them leading to the range of these values.
3. A generalized (with the help of the ranges) digital model of the nature of variation of each parameter value in the ring buffer is created, and written in the appropriate format.
4. This model is sent to the statistical algorithm collecting these models.
5. We loop through the database containing models (signatures) of the nature of changes of values of our parameters and find the model that best suits the current situation.
6. A decision is made as to the system's behavior in the situation captured in this signature (model).
I will formulate it more precisely later.
Unfortunately you are not thinking in the right direction :(
You asked to state the essence, in two words, what I did in vain, though I knew that it is senseless and you immediately engaged in projections of this "figurative" vision on your knowledge base, that unfortunately will not lead to anything good. It is the same as if you made me state in two words the essence of matanalysis and I would say on the example of time series, that the derivative is simply the difference of two neighboring values of the series, and the integral is the cumulative sum and you would immediately rush to inge the Navier-Stokes equations for hydrodynamics. The situation is about the same. MO is artificial intelligence, it is an impressive science, thanks to MO search engines look for what we need better than we would like, thanks to MO machines diagnose medical diagnoses better than doctors, thanks to MO new elementary particles are found in the hadron collider, MO plays chess better, GO and soon any game in general will be played better. It takes at least 5 years to understand the essence of MO, one of 20 with a technical mindset.
Very commendable that you are not afraid to make mistakes and raring to fight, this is a very good mindset, do not lose it, but at the same time try to objectively assess the amount of work required to master this essentially the most difficult area of knowledge, look lectures, my advice to you is a primer.
WhatI have shown is a kind of clustering but with a teacher
classification
Yes, that's right, that's my mistake)
Unfortunately not in the right direction you reason :(
Let him try, why not, his brain is not stained with dogmas and stereotypes, and what if it is interesting?
Let him try it, why not, his brain is not stained with dogmas and stereotypes, and what if he can do something interesting?
It's not for us and you to decide what he can try or not, we are talking about the basic knowledge in this field, without which there is no point in even talking about it all.
To all those who really study MoD.
Please do not get into useless discussions with people who do not know and do not want to know this area of knowledge. "Pioneers" who declare: "Show me, prove.... And then maybe I'll start learning" on this forum, the sea of people. Without having the slightest idea about the subject they with adolescent maximalism will criticize and prove its uselessness. It is not possible and do not need to convince people in the importance and necessity of any knowledge, if they have not matured to them.
This discussion is not only useless, it is harmful. You feed their egos, elevate their criticism, and encourage them to make more garbage statements. We can't ban irrelevant, stupid posts, but we can and should ignore them.
For anyone who wants to understand machine learning and can't search the internet, here's a hint: start here.
https://ru.wikipedia.org/wiki/Машинное_обучение
http://www.r2d3.us/Наглядное-Введение-в-Теорию-Машинного-Обучения/
http://www.machinelearning.ru/wiki/index.php?title=Machine_Learning
http://datareview.info/article/vse-modeli-mashinnogo-obucheniya-imeyut-svoi-nedostatki/
===============================
This thread is getting too big and unreadable. I suggest to start a new branch "RUserGroup" in which only specific issues of applying machine learning models in MT4/5 terminal in languages which allow to do it without problems will be discussed. I know two (R, Python). Discussions to be held with code provided. Experts with some experience in other languages are also welcome.
We can start with an example of convolutional network in previous posts.
Good luck
Thank you to everyone who commented on my undoubtedly amateurish point of view. I don't deny that this topic is unfamiliar to me. Perhaps the ideas presented yesterday have nothing to do with machine learning, but the forum does not prohibit you from being a nerd and weaving pseudoscientific nonsense while trying to understand a complex and important topic. ) I am not afraid of appearing stupid and ignorant.
I believe that no discussion is useless if it makes people think, go beyond stereotypes, and consider new approaches. It's equally useful for beginners and seasoned professionals alike. The only thing I show maximalism in is my aversion to this position:
"We will sit on the MQL forum, promote other languages as more advanced and urge the use of crutch methods in solving problems that MQL fails to cope with. We're not going to implement new features in MQL, helping it grow-we'd rather be chastising and criticizing it. We know a lot about machine learning, but we are not able to implement it in MQL, and those who want to try are just ignoramuses.
I think this is a dead-end position. It contributes neither to personal growth, nor to the development of the language and platform, and doesn't help anyone who prefers MQL. If the only interest is in other people's implementation, then why are we discussing it here? There is an R forum. What good will this discussion do for the development of MQL, if those who consider themselves experts are not going to implement anything in it? What's more, they dissuade others from doing it.
I am well aware of the fact that a large and complex subject cannot be "split into two", but in the spirit of contradicting the local propagandists of other languages, I will advocate MQL and the possibility of implementing new "unavailable" features in it.
So, a little later on, I'll post my concept of machine learning anyway.
And let them throw rotten tomatoes at me.)
Why are there individual enthusiasts who are rewriting the bicycle on MQL?
By the way,recently someone asked for a simple example with the NS,for some reason no one remembered Reshetov's long-standing workaround. By the way,Reshetov himself,as I understand it,is not at all on µl developing his project now.
https://www.mql5.com/ru/code/10289
https://www.mql5.com/ru/code/16727
https://www.mql5.com/ru/code/1104
If there are people who have already laid some basis for the implementation of machine learning in MQL, then it is even more incomprehensible the desire of some to promote other languages. The only thing you need to do is to keep developing this base.
All of the codes above are remakes of Reshetov's work, about which there were disputes 100500 years ago - whether it is NS or so, homemade crap.
The almost complete absence of such Expert Advisors in the Market is an extra proof of that.
so there are no examples of mql-based NS in the public domain.
and this after 6 years of platform and language development. i think this is more than indicative.
You may become a pioneer)) welcome!