Question on neural network programming - page 4

[plan]

2rip: Any suggestions? ;)

[RIP]

plan:
2rip: Any suggestions? ;)

There are always offers ... We need to understand how adequate the exchange will be :)

[Victor Nikolaev]

Oddly enough, the exchange is never mutually beneficial. So someone has to think. And it's a good thing it's not me.

[Victor Nikolaev]

plan:
2rip: Any suggestions? ;)

We could do something simpler. Everyone puts out their work, and others decide. But no one wants to give their own in the wrong hands, although someone has already done it. So the problem is most often found in something else. But if you come to a common idea, you often get a very good result.

[Dmitrii]

plan:
2rip: Any suggestions? ;)

There is a suggestion to use this configuration as a "community cell":

//+------------------------------------------------------------------+
//| 2-х слойная нейросеть из 3-х нейронов                            |
//| первый слой - два нейрона                                        |
//| второй слой - один нейрон                                        |
//| веса W[0], W[N+1], W[2*N+2] - задают смещение нейронов           |
//| N - число входов                                                 |
//| Число подстраиваемых весов расчитывается как N*2+2+3             |
//+------------------------------------------------------------------+
double neuronet1(double W[],double x[]) 
  {
   int i;
   int N=ArraySize(x);
   if(ArraySize(W)!=N*2+5) {
      Print("Ошибка! Размер массива весов должен быть = ",N*2+5);
      return(0);
   }
   double sum=0.0;
   double Out1=0.0, Out2=0.0;
   //---Neuron 1 ----------------+
   sum=W[0];
   //---
   for(i=1; i<=N; i++) { 
      sum+=W[i]*x[i-1];
   }
   Out1=MathArctan(sum);//th(sum);
   //---Neuron 2 ----------------+
   sum=W[N+1];
   //---
   for(i=N+2; i<=2*N+1; i++) { 
      sum+=W[i]*x[i-(N+2)];
   }
   Out2=MathArctan(sum);//th(sum);
   //---Neuron 3 ----------------+
   return(MathArctan(Out1*W[2*N+3]+Out2*W[2*N+4]+W[2*N+2]));
  }
//-------------------------------------------------------------------------------------------------

[plan]

Vinin:

plan:
2rip: Any suggestions? ;)

We can do it in a simpler way. Everyone puts out their work and the others decide. But no one wants to give their own in the wrong hands, although someone has already done it. So the problem is most often found in something else. But if you come to a common idea, you get very good results.

Yes. Putting it on public display is not a good idea, unless of course it is a grail, of which there is plenty! Although I'm for Open Source, but my work on the diploma later will go and it would not be wise to put them on the network before

[plan]

rip:

plan:
2rip: Any suggestions? ;)

There are always offers ... We need to understand if the exchange will be adequate :)

There is no need and therefore no desire to exchange anything ... But help in some matters is always welcome! My ideas I briefly summarized above. Interesting to hear what you have :)

[Dmitrii]

This is an indicator! A self-made GA can be used as a training tool for the above NS:

[rsi]

klot:

There is a suggestion to use this configuration as a 'community cell':

...

Thanks for the code. Why this particular configuration? Do you have any results with it? And can't you use something faster than arctangence as a sigmoid function?

[Dmitrii]

rsi:

klot:

There is a suggestion to use this configuration as a 'community cell':

...

Thanks for the code. Why this particular configuration? Do you have any results with it? And can't you use something faster than arctangence as a sigmoid function?

Any non-linear function will do. It's an indicator!