FullyMeshedNet.template

group FullyMeshedNet;

getOutputDeclaration(randomsource,randombiases,seed,iterations,inputnodes,outputsize,outputnodes,nodes,activationmode,biases,weights) ::= <<

float bias[<nodes>]={<biases:{bia|<bia>f}; separator=", ">};

float randombias[<nodes>]={<randombiases:{bia|<bia>f}; separator=", ">};

float weight[<nodes>][<nodes>]={<weights:{wei|{<wei:{we|<we>f}; separator=", ">}}; separator=", ">};

float netOutput[<outputsize>];

long long int seed=<seed>LL;

long nextValue(){
   seed=(seed*25214903917LL+11) & ((1 \<\< 48) - 1);
   return (long)(seed \>\> 22);
}

float nextDouble(){
   return ((((long long int)nextValue() \<\< 27) + nextValue()) / (float)(1 \<\< 53));
}

float rand_range(float border){
  float val = (float)(nextDouble()*2.0*border-border);
  return val;
}

float sigmoidActivate(float x) {
	float y=(float)(1.0f / (1.0f + exp(-x)));
	return y;
}

float linearActivate(float x){
	if (x \>= 1)
		{return 1;}
	else {
		if (x \<= 0)
			{return 0;}
		else
			{return x;}
		}
}

Result getStep(float netInput[], long inputsize){
    long i;
	long j;
	float activation [<nodes>];
	for (i=0L; i \< <nodes>L; i=i+1){
		activation[i]=0.0f;
	}
    for (i=0L; i \< inputsize; i=i+1) {
      netOutput[i]=netInput[i];
    }
	for (i=<inputnodes>L; i \< <nodes>L; i=i+1) {
      float sumValue=0.0f;
	  for (j=0L; j \< <nodes>L; j=j+1) {
        sumValue=sumValue+weight[j][i]*netOutput[j];
      }
	  activation[i]=bias[i]+sumValue;
	  if (<randomsource>){
		activation[i]=activation[i]+rand_range(randombias[i]);
	  }
    }
	float outputVector [<outputnodes>];
	for (i = <inputnodes>L; i \< <nodes>L; i=i+1) {
      netOutput[i]=<activationmode>(activation[i]);
    }
	for (i = (<nodes>L - <outputnodes>L); i \< <nodes>L; i=i+1) {
	  j = i - (<nodes>L - <outputnodes>L);
      outputVector[j]=netOutput[i];
    }
	Result r(outputVector,<outputnodes>L);
	return r;
}

Result getOutput(float netInput[],long inputsize){
  long i;
  for (i=0L; i \< <outputsize>L; i=i+1) {
    netOutput[i]=0;
  }
  for (i=0L; i \< <iterations>L - 1; i=i+1) {
    getStep(netInput,inputsize);
  }
  return getStep(netInput,inputsize);
}

>>