cbasedann/training.c

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include "randomness.h"
#include "activations.h"
#include "neurons.h"
#include "network.h"
#include "preprocessing.h"
#include "training.h"



void forward(Network *network, Sample *sample)
{
    Feature *current_feature;
    Neuron *current_neuron, *prev_layer_current_neuron;
    Weight *current_weight;
    int i;
    for(i=0 ; i<network->n_layers ; i++)
    {
        if(i==0) //set first layer neurons' output equal to sample's features
        {
            current_feature = sample->first_feature;
            current_neuron = network->layers_first_neurons[i];
            while(current_neuron != NULL)
            {
                current_neuron->output = current_feature->value;
                current_feature = current_feature->next_feature;
                current_neuron = current_neuron->same_layer_next_neuron;
            }
        }else //when layer not first one, do dot product sum with bias
        {
            current_neuron = network->layers_first_neurons[i];
            while(current_neuron != NULL)
            {
                prev_layer_current_neuron = network->layers_first_neurons[i-1];
                current_neuron->output = current_neuron->bias;
                current_weight = current_neuron->weights;
                while(prev_layer_current_neuron != NULL)
                {
                    current_neuron->output += prev_layer_current_neuron->output*current_weight->value;
                    current_weight = current_weight->next;
                    prev_layer_current_neuron = prev_layer_current_neuron->same_layer_next_neuron;
                }
                current_neuron->output = current_neuron->activation( current_neuron->output ); //apply activation function
                current_neuron = current_neuron->same_layer_next_neuron;
            }
        }
    }
}

void errors_propagate(Network *network, Sample *sample)
{

}

void backpropagate(Network *network, float learning_rate)
{

}