Machine learning in trading: theory, models, practice and algo-trading - page 616
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
There are many algorithms, even more than one would like. For example -
Article from Vladimir -https://www.mql5.com/ru/articles/2029
Article by Alexey -https://habrahabr.ru/company/aligntechnology/blog/303750/
It's a heartfelt piece of writing. Now I will see what is there and how in these articles. The Invariant method, too, let me tell you, is quite interesting. What Reshetov has realized... So let's see...
The law of conservation of energy in action))
No one wants to do anything new if the existing MT-R communication tool works.
My connection to MT is now done without DLL at all, by exchanging TXT files via RAM-Disk. So far there has been no need for a DLL. The exchange speed, >1.5 GBytes/s, is enough for anything, and more will remain. Communicate with anyone, even with R, even with a bald devil. And you don't have to do anything at all. I mean, nothing at all.
Details in the thread --https://www.mql5.com/ru/forum/79922
Do you start the R process directly from the terminal too? Or manually?
I manually run the program in R (or other software). From there they exchange themselves and forever. Are you stressed by the extra keystroke before starting?
Lots of advantages compared to DLL.
I like the existing dll better)
That's for now. Everything flows, everything changes.
I prefer not function calls, but full-fledged exchange, when programs work independently and exchange information in both directions.
In my interperation, exchange via DLL and IP-client-server, this is a development of two-way exchange via files -https://www.mql5.com/ru/blogs/post/671000And such exchange simplifies everything dramatically, including all sorts of future system upgrades.
SZY in the end, I don't care what I write in - Python, C++, C#, R, etc., etc. The interfaces don't need to be changed at all. They fit for everything.
Greetings to the minds of the forum. I should say at once that I have not yet mastered the whole branch. But the question of models and neural networks has recently hit me very hard. Maxim not once said that the correlation is kind of a bad method to determine the model, I'll ignore this because in my case the correlation is above 0.95. The question is a bit different, when you run the model through history you get the same 50/50. And here I got an idea, maybe a neuronet is able to determine at what point the model sells and at what point it buys ...? I would like to hear the opinion of experts who have already tried and are working with a neuronet. I attach pictures of the model, where you can see the difference in forward prices.
The first picture shows price going down, the second one is going up and the third one is flatting.
I would like to add that the model can be assembled even every hour because it consists of a portfolio of currency pairs. Due to the fact that the maximum correlation is chosen, the length of the model may be different. And that is where the question arises, what to do when the model length varies, how to feed such samples into the neural network?
Greetings to the minds of the forum. I should say at once that I have not yet mastered the whole branch. But the question of models and neural networks has recently hit me very hard. Maxim not once said that the correlation is kind of a bad method to determine the model, I'll ignore this because in my case the correlation is above 0.95. The question is a bit different, when you run the model through history you get the same 50/50. And here I got an idea, maybe a neuronet is able to determine at what point the model sells and at what point it buys ...? I would like to hear the opinion of experts who have already tried and are working with a neuronet. I attach pictures of the model where you can see the difference in forward prices.
If I look at the first picture, the price is going down, the second one is going up and the third one is flatting.
The answer to your question: Classification. The buy/sell/pass signals.
As for the images, it is not clear what are the lines.
I would like to add that the model can be assembled even every hour because it consists of a portfolio of currency pairs. Due to the fact that the maximum correlation is chosen, the length of the model may be different. And this is where the question arises, how to deal with model length variation, how to feed such samples into a neural network?
I understand that you are talking about dataset.
The length of dataset affects the quality and speed of learning (rows). The quality of parameters (columns) affects the quality of the prediction