"New Neural" is an Open Source neural network engine project for the MetaTrader 5 platform. - page 27

 
-Alexey-:
Absolutely right. Among dozens of primitive regression line indicators and hundreds of posts about it on both forums, there is not a single one that would correctly build and use it. Already in such a simple thing you can get stuck if you are serious. And you say neural network.
There are plenty of correct and proper implementations. You shouldn't say that.
 
Vinin:
No one needs a starship yet. What is needed is the implementation of a few types of networks. No more than that. Let there be several types of input data rationing (if necessary)

-Alexey-:

Really, how anyone can use something in which he understands almost nothing, and with minimal effort - I can not imagine. Yes, the blondes (may blondes forgive me) drive the car, but they know some of its devices, they know the rules, and they drive it in the city, and they don't go out of town, where there are no services. So "any trader" and even "with minimum effort" and even "space plane" and even with profit is not realistic imho, especially for space plane.

Urain:
Do you know what electricity is? In fact, no one really knows what electricity is. But I know that with electricity you can make coffee for a man, and you can make a man himself (from "The Secret").

Trader doesn't need to know how it works, he needs to know the minimum rules of use, our goal is to make these requirements as simple and intuitive as possible.

1. The structure, the architecture and so on can be discussed until the third coming. In my opinion, it is necessary to set a concrete task (not very complicated, but not very simple either), and then try to solve it within the framework of the project.

You have to do everything considering that the result should be universal and intended for any user (like VISARD, I hate to say it).

2. All that it is possible to realize should be realized at once (here it is necessary to move from a scooter to a starship), if something desires to realize but while there is no possibility (or the decision will not be universal) it should be either rejected or taken out of limits of the project.

 

Okay, so we need a scooter?, and then we bolt Belazov wheels to it, but because they are heavy, you'll need a dozen pedal-twirlers, but no problem, add 10 saddles and pedals more, but there will be inefficiency because people will not twist in sync, so we add synchronizers each 50 pounds, but now again not enough people, bullshit, add more saddles,

oopssssssssssssssssssss frame burst....

Have you ever wondered why MQ started writing MT5 snool, instead of continuing to improve the concept of MT4?

Yes all because in the process of development revealed a lot of blunders in the concept itself, it was not enough "space" and reached its ceiling.

If the concept now do not lay in the super capabilities at the end of the frame will burst.

 
Urain:

Okay, so we need a scooter?, and then we bolt Belazov wheels to it, but because they are heavy, you'll need a dozen pedal-twisters, but no problem, add 10 saddles and pedals more, but there will be inefficiency because people will not twist in sync, so we add synchronizers each 50 pounds, but now again not enough people, bullshit, add more saddles,

oopssssssssss frame burst....

Have you ever wondered why MQ started writing MT5 snool, instead of continuing to improve the concept of MT4?

Yes all because in the process of development revealed a lot of blunders in the concept itself, it was not enough "space" and reached its ceiling.

If we don't add super features to the concept now, the frame will burst at the end.


And the example is not correct and it is not necessary to twist and hold us for blondes too ...))
 
Vinin:
There are quite a few correct and proper implementations. You shouldn't do that.
Correct construction is on each new measurement (line shift) introduction of corresponding residual function, providing criterion of maximal efficiency of estimation of linear dependence of analyzed sample, which (function) depends on law of scatter distribution, which (law of scatter) must also be correctly estimated, which is also a task, etc. I haven't seen any implementation of it. This is a non-trivial task.
 
Mischek:
And the example is not correct and it's not necessary to twist and hold us for blondes too ...))

What is the incorrectness?

Well we'll realize MLP and assemble the project and teach the grid to classify XOR, but the topology is not suitable for the same SLTM

And what, change the topology? it's not certain that the new one will be suitable for MLP.

About blondes I did not say a word.

Long short term memory - Wikipedia, the free encyclopedia
Long short term memory - Wikipedia, the free encyclopedia
  • en.wikipedia.org
Long short term memory (LSTM) is a recurrent neural network (RNN) architecture (an artificial neural network) published1 in 1997 by Sepp Hochreiter and Jürgen Schmidhuber. Like most RNNs, an LSTM network is universal in the sense that given enough network units it can compute anything a conventional computer can compute, provided it has the...
 
Avals:

1) Have you invented anything fundamentally new during this time?

2) And what are your successes in building a profitable NS for trading, if you're evaluating someone's knowledge? :)

3) And you need to understand the specifics of trading, that all inefficiencies are temporary and their properties change. That's why it makes no sense to be so bold to dispose of a number of examples in order to eliminate the retraining. Or do you naively hope to create an NS, which can be trained on any section or on any instrument and it will be robust? :)

1) Yes, many things have been invented. gpwr can (if he wants to) give a primer on the subject.

2) I realize that this question is pure provocation. But congratulations, I fell for it. More than a month ago I resigned from my main, fairly well-paid by today's standards, job, and engaged in auto-trading, using the accumulated knowledge and experimental experience in this field of knowledge. People who are close to me, my friends, respected forum users, are aware of the recent events in my life.

3) More than two years of experience in studying and mastering genetic algorithms, whose initial ideas in the source codes are laid out in my article, allow me to say that the construction of a rabbi system is not only possible but also successfully applied in real, everyday life. And I dare say that it is the systems built on neural technologies and self-organization/self-learning/adaptivity principles that are capable of being robust in the long run in the marketplace.

Генетические алгоритмы - это просто!
Генетические алгоритмы - это просто!
  • 2010.05.25
  • Andrey Dik
  • www.mql5.com
В статье автор расскажет об эволюционных вычислениях с использованием генетического алгоритма собственной реализации. Будет показано на примерах функционирование алгоритма, даны практические рекомендации по его использованию.
 

Urain:

Yes all because in the process of development revealed a lot of blunders in the concept itself, it was not enough "space" and reached its ceiling.

If you don't put super features into the concept now, at the end the frame will burst.


Of course it is necessary to lay them, but it is not necessary to implement them.

Firstly it's necessary to understand what is a brick, then what is an ethch, and only then sculpt a skyscraper from a pile of floors.

In the meantime, the discussion is more like a small apartment building with 16-20 stories. In this case, not that the two floors have not learned how to build, with technology and building materials have not figured out.

 
Vinin:
Maybe I can do it. I don't qualify as a programmer, I don't have two degrees. But I'm good at something.
I'm in favor of your candidacy in conjunction with gpwr's candidacy.
 

My opinion is that enough is enough:

Adaline

Kohonen Nets.

LVQ(Learning Vector Quantization)

RBF networks

MLP

and let the user do the normalization).