How to build a NN-EA in MT4

 

Hi guys,

theres been some discussion around how to create a NN-EA for MT4. I have attached an implementation of a very simple yet powerful backpropagation learning, multilayer perceptron class which I wrote in PHP (for easiest understanding this language was the best choice imho), which should be enough for a "proof of concept". Even though its not MT4-code yet: the languages arent that much different and porting a trained working network to MT4 should be trivial to do. I also decided to use php since mt4 doesnt allow class-declarations yet and offers no debugging at all - things not acceptable for serious development.

The attached example is a able to learn any basic logical function (training samples included) like AND or XOR. For more complex patterns (like an EUR/USD chart), more neurons and hidden processing layers must be added, which is a trivial task if you roughly understand the code.

I published the code in order to encourage the discussion what to use as input data for the NN to learn. Of course historical quotes of the security we want to predict. However other values may be very interesting as well. Like oil/gold prices, or quotes of major indices, etcetc. Just anything that may be capable of correctly modifying the output.

To run the program you just need to download the PHP-language [1] and run the .php file with "php-cgi.exe" - simple stuff.

[1] = http://de.php.net/get/php-5.2.5-win32-installer.msi/from/this/mirror

Best wishes,

Heiko

Files:
neuron.zip  3 kb
 

Hi

NN kohenenn

experts\

NS2-32.dll

KOHENENN v15.mq4

KOHENENN v15.ex4

c:\Koh_2007.def

Files:
kohonen.rar  99 kb
 
xan023:
Hi

NN kohenenn

experts\

NS2-32.dll

KOHENENN v15.mq4

KOHENENN v15.ex4

c:\Koh_2007.def

which cross? which time frame?

thanks

Bolla

 

Thanks for your posting but your kohonen net is a dll - so its basically a blackbox that comes with no source. This doesnt help a novice user to understand the concept of NNs. I wanted to encourage to develop an open NN-EA that all coders here can read and use.

 

very interesting

Please keep this thread going, it is very interesting.

More on KOHONEN MAPS

The Self-Organizing Map, or Kohonen Map, is one of the most widely used neural network algorithms, with thousands of applications covered in the literature. It was one of the strong underlying factors in the popularity of neural networks starting in the early 80's. Currently this method has been included in a large number of commercial and public domain software packages.

PS the dll mentioned before is from WARD systems...not sure if it should be posted here...

 

Eur/usd Tf 1h

 
Files:
kohenenn.jpg  63 kb
 

Неплохо летит кривая

 
 

Thanks for the link.

Why not every tick modeling quality 90%?