Probabilistic neural networks, packages and algorithms for MT4 - page 2

New comment
 
YuraZ:
Vinin:
magiXpert:
Statistica. It's new. If you need it, I'll send you the link. Free ;) What more to learn.

My логин@mail.ru It would be appreciated.


about using different neuro packs! i agree with brtter you need to write your own


I don't want to learn C, and what I did was MQL. It takes too much time to learn. And it is not always possible to predict in advance the outcome of the network in the future. If there are tools available, you have to use them.
 
Vinin:
YuraZ:
Vinin:
MagiXpert:
Statistica. It's new. If you need it, I'll send you the link. Free ;) What more to learn.

My логин@mail.ru It would be appreciated.


about using different neuro packs! i agree with brtter you need to write your own


I don't want to learn C and what I did MQL. It takes too much time to learn. And it is not always possible to predict in advance the outcome of the network in the future. If there are tools available, they must be used.


Why should I learn C, (I know it quite well myself - not perfectly)?

Better wrote in C++ only because he knows it well, and it will be faster than MQL

because testing the network and its adjustment would take a lot of time ... but nevertheless everything was rewritten to MQL4

so the author chose C++ because he knows it + time https://championship.mql5.com/2012/ru/news

i had a friend, a talented programmer. he said the best code editor is not the best one, but the one you know!

he wrote talented programs ... which he edited with a fairly simple editor... even though there were a lot of sophisticated ones around.

his competitors were a department of almost 50 people! they used the most sophisticated technology

the result was unfortunate! the group of 50 people's software was thrown away ...

the colleague's software was used for almost 20 years ...

 
About where to look for programs. Does everyone have donkey? I'm not talking about donkey from Issyk-Kul, I'm talking about eMule. Go into donkey. Type in the name of the program you're looking for and go. I get the latest version of Matlab that way.
About the neural networks. I think the main problem is retraining and retraining them in real time. The market is changing. And what worked yesterday will fail today and will fail tomorrow. Which means networks have to be retrained somehow. Especially in championship conditions, when the expo is left without management for 3 whole months. By the way this is my main complaint about the rules of the championship. In short, men, maybe it is better to discuss this issue?
 
eugenk:
Regarding where to look for programs. Does everyone have Oslo? I'm not talking about donkey from Issyk-Kul, I'm talking about eMule. Go to the donkey. Type in the name of the program you're looking for and go. I get the latest version of Matlab that way.
About the neural networks. I think the main problem is learning and retraining them in real time. The market is changing. And what worked yesterday will fail today and will fail tomorrow. Which means networks have to be retrained somehow. Especially in championship conditions, when the exp is left without control for 3 whole months. By the way, this is my main complaint to the championship rules. In short, men, maybe it is better to discuss this issue?


I've put the exp... I only learned it from the 2007 period.

i.e. just picked the optimum parameters - and so far it's happening

there is no retraining inside!

i am not very familiar with the nS! but my first algorithm, when i got into forex, i used to draw it like this

my first block - start - read indicators - summarize indicator readings - signal - received signal - memorize - work out - based on the result correct indicator parameters - retrain

in fact, this is something like a simple NA - although I have not read anything about the NA

it is necessary to retrain - retrain, I think, and probably only when there are bad results - I may be wrong - this is an intuitive opinion

I also intuitively believe that the most difficult part of the algorithm is the learning unit

 
I completely agree with you, although I don't understand much of the theory of building NS. For example, I do not know how to evaluate the criteria for the system that stopped working, it has lost 15% of the deposit or 50%.
 
YuraZ, so you are in the championship and you're not giving up yet? Give me a link to your fighter! There was a discussion about adapting and retraining once. I suggested to write a simple system. For example by stochastic or crossing two mashes. And then just try to optimise it every day over a not very long period. I don't know if anyone has carried out such an experiment. It's difficult in the tester. And it's even harder on the demo. This was just before I disappeared from here and forex in general for almost a year. So, where to look for it now - I don't know. But the difference between the grids and simply optimized expansions is purely quantitative. The grids simply have more parameters, but they do not have any ideas. By the way, how to deal with complete ineffectiveness of grids, I still don't understand. On the one hand it is not good because ideas save computational resources and input information. On the other hand it is good because a strong idea still needs to be found. So, the question is not so simple. But optimization and reoptimization are common that unites both approaches.
 
Perhaps Better's success lies in creating a self-optimising (self-learning) real-time probabilistic neural network.
 
lovova:
eugenk I fully agree with you, although I understand almost nothing in the theory of building NS. for example, it is not clear to me how to assess the criteria that the system stopped working, it has lost 15% of the deposit or 50%.
this is the main question... Now I'm trying to find an answer to the question what is knowledge in general. My approach is something like this. I know the rule of differentiation of a function from a function and I know how to program under Linux on the QT library. Which of these is more valuable in my knowledge? On one hand, QT knowledge helps me earn my living. On the other hand, QT library may become obsolete, and so may Linux itself. And the rule of differentiation of a function from a function will always remain the same. The question is, how to estimate both my knowledge ? It turns out that talking about knowledge, we implicitly talk about Time and Eternity. Exactly the same thing we are talking about when we talk about Forex. The situation by the way is not new at all. Make an experience. Take a long piece of rubber band and hang a weight on it, like a pendulum. Swing the thing. At certain ratios of rubber band length, elasticity and weight, you'll be surprised. The whole thing is described by the Mathieu equation. It is solved approximately by dividing it into fast and slow parts. Alas, there's no such partition in forex. And all our problems are philosophical questions of proper understanding of Time and proper work with this abstraction... Sorry if I've uploaded. But I'm afraid that until we answer these purely philosophical questions in a working order, we will always fail.
 
Gentlemen!
Learning algorithms for meshes, even self-learning ones, are not hard to find, but it's a trifle compared to the process of preparing the data for learning.
 
renegate:
Gentlemen!
Algorithms for training grids, even self-learning ones, are not hard to find, but it's a trifle compared to the process of preparing data for training.


I would clarify. The formulation of the problem. Since everything depends on it.

As I understand it, most of the EAs in the competition have neuronics.

123456789...20
New comment