refactor: re-wrote feed and FeedForeward to run the opposite direction for speed.

Author: gramcracker

header/net.h

@@ -108,7 +108,8 @@ class net //represents a the vector of vectors of neurons
 
   void AddToSequence(int iteration, int layer, int first, int last, void(neuron::*func)());
  // void AddToSequence(int iteration, int layer, void(neuron::*func)());
-
+  //TODO: make main way to run any training and prediction. 
+  //(higher level functions like restricted boltzman set up sequence, then call run sequence)
   void RunSequence();
   ofstream& Save(ofstream& file, bool save_neuron_values =false);
   ifstream& Open(ifstream& file);

src/data_operations.cpp

@@ -9,7 +9,7 @@ using namespace std;
 double data_operations::random_number_generator::generate(){
     mt19937 gen((random_device())());
     uniform_real_distribution<> dis(-1.0, 1.0);
-    return dis(gen);
+        return dis(gen);
     }
 
 //todo:make these use random number generator

src/net.cpp

@@ -328,15 +328,11 @@ void net::FeedForeward(neuron::ActivationType activation_type)
     vector<thread> threads;
     for(int i=0; i<layers.size(); i++)
     {
-        // for(int l=0; l<layers[i].size(); l++)
-        // {
-        //     (((layers[i][l])->_value)=(layers[i][l])->_buffer);//to avoid accessing and activating a neuron at the same time ( like if a neuron is connected to one in the same layer)
-        // }
 
-        if (i==0)_activation_type = neuron::none;//if input layer, dont activate, just multiply by weights
-        else{
-            _activation_type = activation_type;
-        }
+        // if (i==0)_activation_type = neuron::none;//if input layer, dont activate, just multiply by weights
+        // else{
+     _activation_type = activation_type;
+        // }
 
         for(int j=0; j<layers[i].size(); j++)
         {
@@ -743,7 +739,7 @@ net* net::AppendNet(net* net_from, bool pasteOver)
 void net:: Connect(int layer_from, int column_from, int layer_to, int column_to, float weight)
 {
     layers.at(layer_from).at(column_from)->inweights.push_back(new link((*(layers.at(layer_to).at(column_to))), weight));
-    cout<<"new link from "<<layer_from<<":"<<column_from<<" to "<<layer_to<<":"<<column_to<<endl;
+    //cout<<"new link from "<<layer_from<<":"<<column_from<<" to "<<layer_to<<":"<<column_to<<endl;
 }
 
 void net::Connect(int layer_from, int column_from, int layer_to, int column_to)

src/neuron.cpp

@@ -13,7 +13,7 @@ using namespace std;
  */
 neuron::~neuron()
 {
-    cout<<"deleting: "<<this<<endl;
+    //cout<<"deleting: "<<this<<endl;
     int j=0;
         for (  vector <link*>::iterator i=inweights.begin (); i !=inweights.end(); i++)
         {
@@ -43,6 +43,7 @@ void neuron::setin(double input)
  */
 void neuron::activate(ActivationType activationtype)// activation function using a specific Activation type
 {
+    _buffer+=_bias;
     switch(activationtype){
     case none:
         break;
@@ -193,7 +194,7 @@ void neuron::clear()
     _buffer=0;
 }
 
-
+//Todo: switch origin name to destination
 void neuron::feed(ActivationType activationtype){
 
     activate(activationtype); 
@@ -208,6 +209,7 @@ void neuron::feed(ActivationType activationtype){
 }
 
 
+//TODO rename this
 /**
  * @brief      to get the error and pass it back
  */
@@ -229,12 +231,6 @@ void neuron::throwup()
         }
     }
 }
-/*
-virtual void throwup(net::LearningType leaning_type)
-{
-
-}
-*/