Discussion of article "Experiments with neural networks (Part 1): Revisiting geometry"

[MetaQuotes]

New article Experiments with neural networks (Part 1): Revisiting geometry has been published:

In this article, I will use experimentation and non-standard approaches to develop a profitable trading system and check whether neural networks can be of any help for traders.

Optimization and forward test results.

Forward test date from 2021.05.31 to 2022.05.31. Out of all the results, we should choose the one featuring the largest profit factor with the maximum of the complex criterion exceeding 20-40.

Author: Roman Poshtar

[Andrei Bayakou]

I tried neural networks based on the Encog C# library, but I personally did not get anything clear

[Roman Poshtar]

Andrei Bayakou #:
I have tried neural networks based on Encog C# library, but personally I did not get anything clear

Send me a link to the material. I'll take a look.

[Denis Kirichenko]

Cool idea, it's hard to find another word for it ))

If it is a perceptron, where is the activation function? Or I didn't find it?

Then, let's take the code for example:

//+------------------------------------------------------------------+
//| The PERCEPRRON - a perceiving and recognising function |
//+------------------------------------------------------------------+
double perceptron1() 
  {
   double w1 = x1 - 100.0;
   double w2 = x2 - 100.0;
   double w3 = x3 - 100.0;
   double w4 = x4 - 100.0;
   
   double a1 = (ind_In1[1]-ind_In2[1])/PointS1;
   double a2 = (ind_In1[4]-ind_In2[4])/PointS1;
   double a3 = (ind_In1[7]-ind_In2[7])/PointS1;
   double a4 = (ind_In1[10]-ind_In2[10])/PointS1;
   return (w1 * a1 + w2 * a2 + w3 * a3 + w4 * a4);
  }

I realise that the author of this approach is probably not you, but what is the practical sense in subtracting 100,0?

One more remark. Perhaps the network is trained before and not after. In its current form it is just some selection of coefficients with the help of Strategy Tester, not optimisation of weights with the help of any learning method.

[Roman Poshtar]

Denis Kirichenko Strategy Tester, not optimisation of weights with the help of any learning method.

I got the perceptron code itself here https://www.mql5.com/en/code/7917. The shapes and angles are my idea.

[Denis Kirichenko]

Roman Poshtar #:

I got the perceptron code itself here https://www.mql5.com/en/code/7917

Okay. Well, it's not a perceptron. Here is a whole article about the perceptron. Alglib even has a perceptron class: %MQL5\Include\Math\Alglib\dataanalysis.mqh.

It's just that the term "neural networks" is in the title of the article....

[Roman Poshtar]

Denis Kirichenko #:

Right. Well, it's not a perceptron. There is a whole article on the perceptron here. There is even a class of perceptron in Alglib: %MQL5\Include\Math\Alglib\dataanalysis.mqh

It's just that the term "neural networks" is in the title of the article....

We'll look into it. Thanks.

[Сергей Криушин]

Looks to me like a great find everything, and thanks a lot to the author, very good and clear beginning.... looking forward to the continuation...

[Roman Poshtar]

Сергей Криушин #:
Looks to me like a great find everything, and thanks a lot to the author, very good and clear beginning.... Looking forward to the continuation...

Thanks for your review. Very glad it helped.

[adicahyanto]

Nice insight. I also try experimenting with neural network and have come across similar idea, not passing the prices directly to the network as prices have no valid boundaries thus I choose to pass the angle/slope of the moving averages.

Let me know if you found some progress.

[Juan Carlos Lopez Mendez]

I use geometric polyhedra and they are great, pyramid shapes, cubes but there is no programming, it's a completely visual system drawing purely trend lines between highs and lows!!!...I think spatial geometry applies very well to the market...Cheers!!!