Avatar Olexandr
500008

Kub (EURUSD 1 minute)
 
Balance:
130 475.45  
Profit:
0.00  
Equity: 130 475.45  
1

http://forex-pamm.com

Skype: topchilo

ICQ: 462964395

  The trading system is based on neural nets
comments: 
Totals: 708
To add comments, please,log in or register
 
Better wrote:
bbaass писал(а):
Dear Mr. Better ,

Congratulations for your Excellent EA, 

I have some questions about your EA.

1. Is there any relations between your EA and your ICQ Number?

http://championship.mql4.com/2007/users/Better/journal/223

Of course, ICQ number is the main part of the EA!

Have you read that the EA has an artificial intellect? Yes, it has.

Sometimes the EA gets in touch via ICQ with advisors of Trishe and Bernanke and asks them what to do - buy or sell.

And when the market is quiet, the EA browses various forums and sometimes answers strange questions.

------------------------

Sincerely yours,

Kub

your EA get the commands from outside,

so I can't undestand what is the duty of PNN here?


By the way I respect to your EA's result,but

I am sure that External inputs are banned,

as you can see in page:

http://championship.mql4.com/2007/rules


2007.12.19 00:09

GA for optimization:...

Max{NNProfit}, where NNProfit=f(Long Patter Ema i , Ema i(period),Nro Neurons Hideen Layer)....FOR MLP

For PNN is more easy,,,inclusive,,,yo could use metatrader optimizer ( have a GA in the Optimizer) like a unsupervized training...

Look 'Artificial Intellegence' Reshetov's work, he uses GA from optimizer to train a simple layer perceptron to get wij ( like unsupervised learning) , that´s a "different" way to train..... the clasical way in this case is delta rule. .. .

my englishhh es "maloooooo"...

Saludos

:)

2007.12.18 22:11

Better wrote:
And at last, but not least: how did you selected the input parameters? Have you used some genetic algorithms od just plain observation?
I haven't used GA, only plain observations

That's very impressive.


Spasiba.
2007.12.18 21:36
emsi писал(а):
 
I understad that both the indicators you use and the NN itself are to be secret but I'm interested if you could reveal the rough number of inputs to your network.
No
Did you used a time-series approach? I understand that the output is only one: a value in <-1;1> and you split it into 3 parts meaning sell, dont-know and buy. You wrote that it's a modified PN so I guess that it's splitted like in this work: http://www.mini.pw.edu.pl/~jaruszewicz/FIN-PRE-MJR-doktorat-3v02.pdf or is it completly different? I mean grouping some inputs in a sub NN. 
I havent read this paper. Can't read in Polish.

Do you have any stats about your 3 sub-EA forecasting accuracy? I mean how many times there was forecast to buy while actually there shoud be sell or vice versa?
On the Reports page one can see that the success rate for long, and short is about 2/3 (66.31 and 66.49 percent) with amazing accuracy. It can't be a coincidence! How does it look fore each of your sub-EA?
You can calculate it by yourself on a basis of my statement. Three subsystems have different magic numbers and indicies [0], [1] and [2] in comments. The detailed statement can be produced in MetatTrader terminal.

And at last, but not least: how did you selected the input parameters? Have you used some genetic algorithms od just plain observation?
I haven't used GA, only plain observations
A bonus question :)
You wrote that you are a trader. If so: if your EA were revealed to the public what would be the impact to the market if any?
I don't know
1
2007.12.18 21:06

emsi wrote:
Hello!

First of all I would like to congratulate you the leading of the championship!

I'm starting a PhD course now and it's about an automated trading systems, including (but not limited to) neural networks approach and I found your EA extremly interesting due to its successfullness.
There's been a number of works in this subject, but none has proofed to work with such efficiency.
I understad that both the indicators you use and the NN itself are to be secret but I'm interested if you could reveal the rough number of inputs to your network. Is it among 4 to 8, 8 to 15, 15 to 30 or more? Did you used a time-series approach? I understand that the output is only one: a value in <-1;1> and you split it into 3 parts meaning sell, dont-know and buy. You wrote that it's a modified PN so I guess that it's splitted like in this work: http://www.mini.pw.edu.pl/~jaruszewicz/FIN-PRE-MJR-doktorat-3v02.pdf or is it completly different? I mean grouping some inputs in a sub NN.

Do you have any stats about your 3 sub-EA forecasting accuracy? I mean how many times there was forecast to buy while actually there shoud be sell or vice versa?
On the Reports page one can see that the success rate for long, and short is about 2/3 (66.31 and 66.49 percent) with amazing accuracy. It can't be a coincidence! How does it look fore each of your sub-EA?

And at last, but not least: how did you selected the input parameters? Have you used some genetic algorithms od just plain observation?

A bonus question :)
You wrote that you are a trader. If so: if your EA were revealed to the public what would be the impact to the market if any?
Is there any chance I could get a reply? I would be very thankfull.
Regards.
2007.12.18 18:44
:) 
;)
 
2007.12.18 15:16

Уважаемый Александр !

Безусловно, Вы уже легенда науки о Форексе и иных рынках, - Ваш эксперт делает невозможное. Правительству Украины можно было бы позавидовать, если бы Вы применили свои разработки в государственных целях ;-) Но не могли бы Вы и нам указать хотя бы "набор средних", тот самый, на котором обучется нейросеть? Может и мы обучимся, уважаемый kub :)

2007.12.18 14:16
不知说什么~~哈哈..还是说声你好厉害啊~~
2007.12.18 10:51
+1
2007.12.18 10:38

rodragon wrote:

La idea de los foros es discutir de Forex, pero aqui todos preguntan y no hay respuestas, yo les voy a contar como funciona este tema (jejej):

...


Saludos a todos, espero que ahora exista un poco mas de luz en este tema.





I agree. Too many questions and no answers. In the forums we are supposed to learn things or show others how things are done. I don't mean revealing your systems to others. However, one can show the workings of a PNN, or any NN, by showing generic examples like the ones you find in text books. Contestants should know that whatever control they think they have over their creation is just a mirage. This control was lost when control was relinquished to a third party who, btw, may or may not honor his/her promise to protect/use your creation. The only thing that can protect your system is security code, or self destructing code, that disables the EA after the contest is over. I stopped asking questions long ago. I went out and bought some books...Anybody interested in learning how NNs or Artificial Intelligence work should do the same because few people in here are willing to share zilch! There are plenty of book out there written by goodhearted researchers willing to show their findings with others... Sorry for being so blunt but i call it the way i see it...Good luck to all and happy trading!

Rodragon you are a good soul for trying to shed some light into the subject. Some day if you stumble upon an excellent EA, don't give it away, put your findings in a Trading For Dummies book and make it a best seller :))
2007.12.18 09:52
comments: