Club Amiga de Montreal - CAM/CAM_CD_1.iso/files/618a.lha/NeuralNetwork/neural_net.c

C/C++ Source or Header | 1992-03-08 | 31KB | 1,012 lines

#include <stdio.h>
#include <math.h>
#include <stdlib.h>
#include "Neural_net.h"

static void Neural_net_allocate_matrices (Neural_net *nn);
static void Neural_net_initialize_matrices (Neural_net *nn,
                                           double range);
static void Neural_net_deallocate_matrices (Neural_net *nn);

static void Neural_net_allocate_matrices (Neural_net *nn)
{

  /* Time to allocate the entire neural_net structure
  // Activation matrices */
  nn->hidden1_act = (double *) malloc (sizeof (double) * nn->num_hidden1);
  nn->hidden2_act = (double *) malloc (sizeof (double) * nn->num_hidden2);
  nn->output_act = (double *) malloc (sizeof (double) * nn->num_outputs);

  /* Weight matrices */
  nn->input_weights = (double *) malloc (sizeof (double) * nn->num_hidden1 *
                                         nn->num_inputs);
  nn->hidden1_weights = (double *) malloc (sizeof (double) * nn->num_hidden2 *
                                           nn->num_hidden1);
  nn->hidden2_weights = (double *) malloc (sizeof (double) * nn->num_outputs *
                                           nn->num_hidden2);

  /* Learning rate matrices for each weight's learning rate.
  // Needed for delta-bar-delta algorithm */
  nn->input_learning_rate = (double *) malloc (sizeof (double) *
                                               nn->num_hidden1 * nn->num_inputs);
  nn->hidden1_learning_rate = (double *) malloc (sizeof (double) *
                                                 nn->num_hidden2 * nn->num_hidden1);
  nn->hidden2_learning_rate = (double *) malloc (sizeof (double) *
                                                 nn->num_outputs * nn->num_hidden2);

  /* Learning rate deltas for each weight's learning rate.
  // Needed for delta-bar-delta algorithm. */
  nn->input_learning_delta = (double *) malloc (sizeof (double) *
                                                nn->num_hidden1 * nn->num_inputs);
  nn->hidden1_learning_delta = (double *) malloc (sizeof (double) *
                                                  nn->num_hidden2 * nn->num_hidden1);
  nn->hidden2_learning_delta = (double *) malloc (sizeof (double) *
                                                  nn->num_outputs * nn->num_hidden2);

  /* Delta bar matrices for each weight's delta bar.
  // Needed for delta-bar-delta algorithm. */
  nn->input_learning_delta_bar = (double *) malloc (sizeof (double) *
                                                    nn->num_hidden1 * nn->num_inputs);
  nn->hidden1_learning_delta_bar = (double *) malloc (sizeof (double) *
                                                   nn->num_hidden2 * nn->num_hidden1);
  nn->hidden2_learning_delta_bar = (double *) malloc (sizeof (double) *
                                                   nn->num_outputs * nn->num_hidden2);

  /* Weight delta matrices for each weights delta.
  // Needed for BackPropagation algorithm. */
  nn->input_weights_sum_delta = (double *) malloc (sizeof (double) *
                                                   nn->num_hidden1 * nn->num_inputs);
  nn->hidden1_weights_sum_delta = (double *) malloc (sizeof (double) *
                                                   nn->num_hidden2 * nn->num_hidden1);
  nn->hidden2_weights_sum_delta = (double *) malloc (sizeof (double) *
                                                   nn->num_outputs * nn->num_hidden2);

  /* Sum of delta * weight matrices for each weight.
  // Needed for BackPropagation algorithm. */
  nn->hidden1_sum_delta_weight = (double *) malloc (sizeof (double) *
                                                    nn->num_hidden2 * nn->num_hidden1);
  nn->hidden2_sum_delta_weight = (double *) malloc (sizeof (double) *
                                                    nn->num_outputs * nn->num_hidden2);

  /* Done neural net allocation */
}

static void Neural_net_initialize_matrices (Neural_net *nn,
                                            double range)
{
  int    x,y;
  double rand_max;

  nn->training_examples = 0;
  nn->examples_since_update = 0;

  rand_max = RAND_MAX;
  /* Initialize all weights from -range to +range randomly */
  for (x = 0; x < nn->num_hidden1; ++x)
    {
      nn->hidden1_sum_delta_weight [x] = 0.0;

      for (y = 0; y < nn->num_inputs; ++y)
        {
          nn->input_weights [x * nn->num_inputs + y] = rand () / rand_max * range;
          if ( rand () < (RAND_MAX / 2) )
              nn->input_weights [x * nn->num_inputs + y] *= -1.0;

          nn->input_weights_sum_delta [x * nn->num_inputs + y] = 0.0;
          nn->input_learning_rate [x * nn->num_inputs + y] = nn->learning_rate;
          nn->input_learning_delta [x * nn->num_inputs + y] = 0.0;
          nn->input_learning_delta_bar [x * nn->num_inputs + y] = 0.0;
        }
    }

  for (x = 0; x < nn->num_hidden2; ++x)
    {
      nn->hidden2_sum_delta_weight [x] = 0.0;

      for (y = 0; y < nn->num_hidden1; ++y)
        {
          nn->hidden1_weights [x * nn->num_hidden1 + y] = rand () / rand_max * range;
          if ( rand () < (RAND_MAX / 2) )
              nn->hidden1_weights [x * nn->num_hidden1 + y] *= -1.0;

          nn->hidden1_weights_sum_delta [x * nn->num_hidden1 + y] = 0.0;
          nn->hidden1_learning_rate [x * nn->num_hidden1 + y] = nn->learning_rate;
          nn->hidden1_learning_delta [x * nn->num_hidden1 + y] = 0.0;
          nn->hidden1_learning_delta_bar [x * nn->num_hidden1 + y] = 0.0;
        }
    }

  for (x = 0; x < nn->num_outputs; ++x)
    {
      for (y = 0; y < nn->num_hidden2; ++y)
        {
          nn->hidden2_weights [x * nn->num_hidden2 + y] = rand () / rand_max * range;
          if ( rand () < (RAND_MAX / 2) )
              nn->hidden2_weights [x * nn->num_hidden2 + y] *= -1.0;

          nn->hidden2_weights_sum_delta [x * nn->num_hidden2 + y] = 0.0;
          nn->hidden2_learning_rate [x * nn->num_hidden2 + y] = nn->learning_rate;
          nn->hidden2_learning_delta [x * nn->num_hidden2 + y] = 0.0;
          nn->hidden2_learning_delta_bar [x * nn->num_hidden2 + y] = 0.0;
        }
    }

}

static void Neural_net_deallocate_matrices (Neural_net *nn)
{

  if ( nn->hidden1_act == NULL )
      return;

  /* Time to destroy the entire neural_net structure
  // Activation matrices */
  free (nn->hidden1_act);
  free (nn->hidden2_act);
  free (nn->output_act);

  /* Weight matrices */
  free (nn->input_weights);
  free (nn->hidden1_weights);
  free (nn->hidden2_weights);

  /* Learning rate matrices for each weight's learning rate.
  // Needed for delta-bar-delta algorithm */
  free (nn->input_learning_rate);
  free (nn->hidden1_learning_rate);
  free (nn->hidden2_learning_rate);

  /* Learning rate deltas for each weight's learning rate.
  // Needed for delta-bar-delta algorithm. */
  free (nn->input_learning_delta);
  free (nn->hidden1_learning_delta);
  free (nn->hidden2_learning_delta);

  /* Delta bar matrices for each weight's delta bar.
  // Needed for delta-bar-delta algorithm. */
  free (nn->input_learning_delta_bar);
  free (nn->hidden1_learning_delta_bar);
  free (nn->hidden2_learning_delta_bar);

  /* Weight delta matrices for each weights delta.
  // Needed for BackPropagation algorithm. */
  free (nn->input_weights_sum_delta);
  free (nn->hidden1_weights_sum_delta);
  free (nn->hidden2_weights_sum_delta);

  /* Sum of delta * weight matrices for each weight.
  // Needed for BackPropagation algorithm. */
  free (nn->hidden1_sum_delta_weight);
  free (nn->hidden2_sum_delta_weight);

  /* Done neural net deallocation */
}

Neural_net *Neural_net_constr (int number_inputs, int number_hidden1,
                           int number_hidden2, int number_outputs,
                           double t_epsilon, double t_skip_epsilon,
                           double t_learning_rate, double t_theta,
                           double t_phi, double t_K, double range)
{
  Neural_net *nn;

  if ( (nn = malloc (sizeof (Neural_net))) == NULL )
    {
      return (NULL);
    }

  nn->num_inputs = number_inputs;
  nn->num_hidden1 = number_hidden1;
  nn->num_hidden2 = number_hidden2;
  nn->num_outputs = number_outputs;
  nn->epsilon = t_epsilon;
  nn->skip_epsilon = t_skip_epsilon;
  nn->learning_rate = t_learning_rate;
  nn->theta = t_theta;
  nn->phi = t_phi;
  nn->K = t_K;
  nn->training_examples = 0;
  nn->examples_since_update = 0;

  Neural_net_allocate_matrices (nn);
  Neural_net_initialize_matrices (nn,range);

  return (nn);
}


Neural_net *Neural_net_read_constr (char *filename, int *file_error,
                                double t_epsilon, double t_skip_epsilon,
                                double t_learning_rate, double t_theta,
                                double t_phi, double t_K)
{
  Neural_net *nn;

  if ( (nn = malloc (sizeof (Neural_net))) == NULL )
    {
      return (NULL);
    }

  nn->epsilon = t_epsilon;
  nn->skip_epsilon = t_skip_epsilon;
  nn->learning_rate = t_learning_rate;
  nn->theta = t_theta;
  nn->phi = t_phi;
  nn->K = t_K;
  nn->training_examples = 0;
  nn->examples_since_update = 0;

  nn->hidden1_act = NULL;
  if ( (*file_error = Neural_net_read_weights (nn,filename)) < 0 )
    {
      free (nn);
      return (NULL);
    }

  return (nn);
}

#ifndef __inline__
Neural_net *Neural_net_default_constr (int number_inputs, int number_hidden1,
                                       int number_hidden2, int number_outputs)
{
  return (Neural_net_constr (number_inputs,number_hidden1,number_hidden2,
                             number_outputs,0.1,0.05,0.1,1.0,0.0,0.0,1.0));
}

Neural_net *Neural_net_default_read_constr (char *filename, int *file_error)
{
  return (Neural_net_read_constr (filename,file_error,0.1,0.05,0.1,1.0,0.0,
                                  0.0));
}
#endif


void Neural_net_destr (Neural_net *nn)
{
  if ( nn != NULL )
    {
      Neural_net_deallocate_matrices (nn);
      free (nn);
    }
}

int Neural_net_read_weights (Neural_net *nn, char *filename)
{
  FILE    *fp;
  int     x;
  long    iter;

  if ( ((fp = fopen (filename,"rt")) == NULL) )
    {
      printf ("Could not read weights from file '%s'\n",filename);
      return (-1);
    }

  /* First read in how many iterations have been performed so far */
  fscanf (fp,"%ld",&iter);
  printf ("Iterations = %ld\n",iter);

  /* Next read in how many input nodes, hidden1 nodes, hidden2 nodes */
  /* and output nodes. */
  fscanf (fp,"%d%d%d%d",&nn->num_inputs,&nn->num_hidden1,&nn->num_hidden2,
                        &nn->num_outputs);

  /* Deallocate previous matrices */
  Neural_net_deallocate_matrices (nn);

  /* Allocate new matrices with new size */
  Neural_net_allocate_matrices (nn);

  /* Initialize all matrices and variables */
  Neural_net_initialize_matrices (nn,1.0);
  nn->training_examples = iter;

  /* Read input->hidden1 weights from file. */
  for (x = 0; x < nn->num_inputs * nn->num_hidden1; ++x)
    {
      fscanf (fp,"%lf",&nn->input_weights [x]);
    }

  /* Read hidden1->hidden2 weights from file. */
  for (x = 0; x < nn->num_hidden1 * nn->num_hidden2; ++x)
    {
      fscanf (fp,"%lf",&nn->hidden1_weights [x]);
    }

  /* Read hidden2->output weights from file. */
  for (x = 0; x < (nn->num_hidden2 * nn->num_outputs); ++x)
    {
      fscanf (fp,"%lf",&nn->hidden2_weights [x]);
    }

  /* Now all the weights have been loaded */
 fclose (fp);

 return (0);
}



int Neural_net_save_weights (Neural_net *nn, char *filename)
{
  FILE *fp;
  int   x;

  if ( ((fp = fopen (filename,"wt")) == NULL) )
    {
      printf ("Could not save weights to file '%s'\n",filename);
      return (-1);
    }

  /* First write out how many iterations have been performed so far */
  fprintf (fp," %ld\n",nn->training_examples);

  /* Next write out how many input nodes, hidden1 nodes, hidden2 nodes */
  /* and output nodes. */
  fprintf (fp," %d %d %d %d\n",nn->num_inputs,nn->num_hidden1,
           nn->num_hidden2,nn->num_outputs);

  /* Write input->hidden1 weights to output. */
  for (x = 0; x < nn->num_inputs * nn->num_hidden1; ++x)
    {
      fprintf (fp,"%10.5f ",nn->input_weights [x]);
      if ( (x % 5) == 4 )
          fprintf (fp,"\n");
    }
  fprintf (fp,"\n\n");

  /* Write hidden1->hidden2 weights to output. */
  for (x = 0; x < nn->num_hidden1 * nn->num_hidden2; ++x)
    {
      fprintf (fp,"%10.5f ",nn->hidden1_weights [x]);
      if ( (x % 5) == 4 )
          fprintf (fp,"\n");
    }
  fprintf (fp,"\n\n");

  /* Write hidden2->output weights to output. */
  for (x = 0; x < (nn->num_hidden2 * nn->num_outputs); ++x)
    {
      fprintf (fp,"%10.5f ",nn->hidden2_weights [x]);
      if ( (x % 5) == 4 )
          fprintf (fp,"\n");
    }
  fprintf (fp,"\n\n");

  fflush (fp);
  printf ("Flushed file\n");
  fclose (fp);
  printf ("Closed file\n");

  /* Now all the weights have been saved */
  return (0);
}


#ifndef __inline__
double Neural_net_get_hidden1_activation (Neural_net *nn, int node)
{
  return (nn->hidden1_act [node]);
}

double Neural_net_get_hidden2_activation (Neural_net *nn, int node)
{
  return (nn->hidden2_act [node]);
}

double Neural_net_get_output_activation (Neural_net *nn, int node)
{
  return (nn->output_act [node]);
}

double Neural_net_get_input_weight (Neural_net *nn, int input_node,
                                    int hidden1_node)
{
  return (nn->input_weights [hidden1_node * nn->num_inputs + input_node]);
}

double Neural_net_get_hidden1_weight (Neural_net *nn, int hidden1_node,
                                      int hidden2_node)
{
  return (nn->hidden1_weights [hidden2_node * nn->num_hidden1 +
                               hidden1_node]);
}

double Neural_net_get_hidden2_weight (Neural_net *nn, int hidden2_node,
                                      int output_node)
{
  return (nn->hidden2_weights [output_node*nn->num_hidden2+hidden2_node]);
};

#endif

void Neural_net_set_size_parameters (Neural_net *nn, int number_inputs,
                                     int number_hidden1, int number_hidden2,
                                     int number_outputs, double range)
{
  double *new_input_weights,*new_hidden1_weights;
  double *new_hidden2_weights;
  double rand_max;
  int    x;

  rand_max = RAND_MAX;

  /* Allocate new weight matrices with new size */
  new_input_weights = (double *) malloc (sizeof (double) *
                                     number_hidden1 * number_inputs);
  new_hidden1_weights = (double *) malloc (sizeof (double) *
                                       number_hidden2 * number_hidden1);
  new_hidden2_weights = (double *) malloc (sizeof (double) *
                                       number_outputs * number_hidden2);

  /* Copy over all weights
  // Input weights */
  for (x = 0; x < number_hidden1 * number_inputs; ++x)
    {
      /* IF the new size is larger than the old size, THEN make new connections
      // a random weight between +-range. */
      if ( x >= (nn->num_hidden1 * nn->num_inputs) )
        {
          new_input_weights [x] = rand () / rand_max * range;
          if ( rand () < (RAND_MAX / 2) )
              new_input_weights [x] *= -1.0;
        }
      else
          new_input_weights [x] = nn->input_weights [x];
    }

  /* Hidden1 weights */
  for (x = 0; x < number_hidden2 * number_hidden1; ++x)
    {
      /* IF the new size is larger than the old size, THEN make new connections
      // a random weight between +-range. */
      if ( x >= (nn->num_hidden2 * nn->num_hidden1) )
        {
          new_hidden1_weights [x] = rand () / rand_max * range;
          if ( rand () < (RAND_MAX / 2) )
              new_hidden1_weights [x] *= -1.0;
        }
      else
          new_hidden1_weights [x] = nn->hidden1_weights [x];
    }

  /* Hidden2 weights */
  for (x = 0; x < number_outputs * number_hidden2; ++x)
    {
      /* IF the new size is larger than the old size, THEN make new connections
      // a random weight between +-range. */
      if ( x >= (nn->num_outputs * nn->num_hidden2) )
        {
          new_hidden2_weights [x] = rand () / rand_max * range;
          if ( rand () < (RAND_MAX / 2) )
              new_hidden2_weights [x] *= -1.0;
        }
      else
          new_hidden2_weights [x] = nn->hidden2_weights [x];
    }

  /* All weights have been copied. */

  /* Change size paramters */
  nn->num_inputs = number_inputs;
  nn->num_hidden1 = number_hidden1;
  nn->num_hidden2 = number_hidden2;
  nn->num_outputs = number_outputs;

  /* Deallocate all matrices */
  Neural_net_deallocate_matrices (nn);

  /* Allocate new nerual network matrices with the correct size and initialize */
  Neural_net_allocate_matrices (nn);
  Neural_net_initialize_matrices (nn,1.0);

  /* Now deallocate new randomly initialized weight matrices and assign them
  // to the new weight matrices that have the correct weight values. */
  free (nn->input_weights);
  free (nn->hidden1_weights);
  free (nn->hidden2_weights);

  nn->input_weights = new_input_weights;
  nn->hidden1_weights = new_hidden1_weights;
  nn->hidden2_weights = new_hidden2_weights;

}

#ifndef __inline__
int Neural_net_get_number_inputs (Neural_net *nn)
{
  return (nn->num_inputs);
}

int Neural_net_get_number_hidden1 (Neural_net *nn)
{
  return (nn->num_hidden1);
}

int Neural_net_get_number_hidden2 (Neural_net *nn)
{
  return (nn->num_hidden2);
}

int Neural_net_get_number_outputs (Neural_net *nn)
{
  return (nn->num_outputs);
}

void Neural_net_set_epsilon (Neural_net *nn, double eps)
{
  nn->epsilon = eps;
}

void Neural_net_set_skip_epsilon (Neural_net *nn, double eps)
{
  nn->skip_epsilon = eps;
}

void Neural_net_set_learning_rate (Neural_net *nn, double l_rate)
{
  nn->learning_rate = l_rate;
}

void Neural_net_set_theta (Neural_net *nn, double t_theta)
{
  nn->theta = t_theta;
}

void Neural_net_set_phi (Neural_net *nn, double t_phi)
{
  nn->phi = t_phi;
}

void Neural_net_set_K (Neural_net *nn, double t_K)
{
  nn->K = t_K;
}

double Neural_net_get_epsilon (Neural_net *nn)
{
  return (nn->epsilon);
}

double Neural_net_get_skip_epsilon (Neural_net *nn)
{
  return (nn->skip_epsilon);
}

double Neural_net_get_learning_rate (Neural_net *nn)
{
  return (nn->learning_rate);
}

double Neural_net_get_theta (Neural_net *nn)
{
  return (nn->theta);
}

double Neural_net_get_phi (Neural_net *nn)
{
  return (nn->phi);
}

double Neural_net_get_K (Neural_net *nn)
{
  return (nn->K);
}

long   Neural_net_get_iterations (Neural_net *nn)
{
  return (nn->training_examples);
}

#endif

void Neural_net_back_propagation (Neural_net *nn, double *input,
                                  double *desired_output, int *done)
{
  int     x,y;
  register int size;
  double  delta,sum_delta,*weight,*p_sum_delta,*p_learning_delta;

  /* First check if training complete. */
  for (x = nn->num_outputs - 1; x >= 0; --x)
    {
      if ( fabs (desired_output [x] - nn->output_act [x]) > 0.1 )
        {
          *done = 0;
        }
    }

  /* Go backward through list for speed */
  size = nn->num_hidden2;
  /* First calculate deltas of weights from output to hidden layer 2. */
  for (x = nn->num_outputs - 1; x >= 0; --x)
    {
      weight = &nn->hidden2_weights [x * size];
      p_sum_delta = &nn->hidden2_weights_sum_delta [x * size];
      p_learning_delta = &nn->hidden2_learning_delta [x * size];
      for (y = nn->num_hidden2 - 1; y >= 0; --y)
        {
          /* Formula delta = (desired - actual) * derivative
             derivative = S(1 - S)
             Also calculate sum of deltas * weight for next layer.
          */
          delta = (desired_output [x] - nn->output_act [x])
                   * SLOPE * nn->output_act [x] * (1.0 - nn->output_act [x]);

          nn->hidden2_sum_delta_weight [y] += delta * weight [y];

          p_learning_delta [y] += delta;

          /* Now multiply by activation and sum in weights sum delta */
          p_sum_delta [y] += delta * nn->hidden2_act [y];
        }
    }


  /* Next calculate deltas of weights between hidden layer2 and hidden
     layer 1 */
  size = nn->num_hidden1;
  for (x = nn->num_hidden2 - 1; x >= 0; --x)
    {
      sum_delta = nn->hidden2_sum_delta_weight [x];
      weight = &nn->hidden1_weights [x * size];
      p_sum_delta = &nn->hidden1_weights_sum_delta [x * size];
      p_learning_delta = &nn->hidden1_learning_delta [x * size];
      for (y = nn->num_hidden1 - 1; y >= 0; --y)
        {
          /* Formula delta = SUM (previous deltas*weight)
                             * derivative
             previous deltas already muliplied by weight.
             derivative = S(1 - S)

             Also calculate sum of deltas * weight to save from doing
             it for next layer.
          */

          delta = sum_delta * nn->hidden2_act [x] *
                  (1.0 - nn->hidden2_act [x]) * SLOPE;

          nn->hidden1_sum_delta_weight [y] += delta * weight [y];

          p_learning_delta [y] += delta;

          /* Now multiply by activation and sum in weights_sum_delta */
          p_sum_delta [y] += delta * nn->hidden1_act [y];
        }
      nn->hidden2_sum_delta_weight [x] = 0.0;
    }

  /* Finally calculate deltas of weights between hidden layer 1 and input
     layer */
  size = nn->num_inputs;
  for (x = nn->num_hidden1 - 1; x >= 0; --x)
    {
      sum_delta = nn->hidden1_sum_delta_weight [x];
      p_sum_delta = &nn->input_weights_sum_delta [x * size];
      p_learning_delta = &nn->input_learning_delta [x * size];
      for (y = nn->num_inputs - 1; y >= 0; --y)
        {
          /* Formula delta = SUM (previous deltas*weight)
                             * derivative * activation of input
             previous deltas already muliplied by weight
             derivative = S(1 - S)
          */
          delta = sum_delta * nn->hidden1_act [x] *
                  (1.0 - nn->hidden1_act [x]) * SLOPE;

          p_learning_delta [y] += delta;

          p_sum_delta [y] += (delta * input [y]);
        }
      nn->hidden1_sum_delta_weight [x] = 0.0;
    }

  /* Now all deltas have been calculated and added into their appropriate
     neuron connection. */
  ++nn->examples_since_update;

}


double Neural_net_calc_forward (Neural_net *nn, double *input,
                                double *desired_output, int *num_wrong,
                                int *skip, int print_it, int *actual_printed)
{
  int     x,y,wrong;
  int     size;
  double  *weight,error,abs_error;

  *skip = 1;
  wrong = 0;
  error = 0.0;

  /* Go backward for faster processing */
  /* Calculate hidden layer 1's activation */
  size = nn->num_inputs;
  for (x = nn->num_hidden1 - 1; x >= 0;  --x)
    {
      nn->hidden1_act [x] = 0.0;
      weight = &nn->input_weights [x * size];
      for (y = nn->num_inputs - 1; y >= 0; --y)
        {
          nn->hidden1_act [x] += (input [y] * weight [y]);
        }
      nn->hidden1_act [x] = S(nn->hidden1_act [x]);
    }

  /* Calculate hidden layer 2's activation */
  size = nn->num_hidden1;
  for (x = nn->num_hidden2 - 1; x >= 0; --x)
    {
      nn->hidden2_act [x] = 0.0;
      weight = &nn->hidden1_weights [x * size];
      for (y = nn->num_hidden1 - 1; y >= 0; --y)
        {
          nn->hidden2_act [x] += (nn->hidden1_act [y] * weight [y]);
        }
      nn->hidden2_act [x] = S(nn->hidden2_act [x]);
    }

  /* Calculate output layer's activation */
  size = nn->num_hidden2;
  for (x = nn->num_outputs - 1; x >= 0; --x)
    {
      nn->output_act [x] = 0.0;
      weight = &nn->hidden2_weights [x * size];
      for (y = nn->num_hidden2 - 1; y >= 0; --y)
        {
          nn->output_act [x] += nn->hidden2_act [y] * weight [y];
        }
      nn->output_act [x] = S(nn->output_act [x]);
      abs_error = fabs (nn->output_act [x] - desired_output [x]);
      error += abs_error;
      if ( abs_error > 0.1 )
          wrong = 1;
      if ( abs_error > nn->epsilon )
          *skip = 0;
    }

  if ( wrong )
      ++(*num_wrong);

  if ( print_it == 2 )
    {
      for (x = 0; x < nn->num_outputs; ++x)
        {
          printf ("%6.4f ",nn->output_act [x]);
        }
      ++(*actual_printed);
    }
  else if ( print_it && wrong )
    {
      for (x = 0; x < nn->num_outputs; ++x)
        {
          printf ("%6.4f ",fabs (desired_output [x] - nn->output_act [x]));
        }
      ++(*actual_printed);
    }

  return (error);

}


void Neural_net_update_weights (Neural_net *nn)
{
  int     x,y;
  int     size;
  double  rate,*p_ldelta,*p_ldelta_bar,*weight,*p_lrate,*p_sum_delta;

  /* Check to see if any changes have been calculated. */
  if ( nn->examples_since_update == 0 )
    {
      return;
    }

  /* Go backward for slightly faster processing */
  /* First add deltas of weights from output to hidden layer 2. */
  size = nn->num_hidden2;
  for (x = nn->num_outputs - 1; x >= 0; --x)
    {
      p_ldelta = &nn->hidden2_learning_delta [x * size];
      p_ldelta_bar = &nn->hidden2_learning_delta_bar [x * size];
      weight = &nn->hidden2_weights [x * size];
      p_lrate = &nn->hidden2_learning_rate [x * size];
      p_sum_delta = &nn->hidden2_weights_sum_delta [x * size];

      for (y = nn->num_hidden2 - 1; y >= 0; --y)
        {
          rate = p_ldelta [y] * p_ldelta_bar [y];
          if ( rate < 0.0 )
            {
              p_lrate [y] -= (nn->phi * p_lrate [y]);
            }
          else if ( rate > 0.0 )
            {
              p_lrate [y] += nn->K;
            }

          weight [y] += (p_lrate [y] * p_sum_delta [y]);
          p_sum_delta [y] = 0.0;
          p_ldelta_bar [y] *= nn->theta;
          p_ldelta_bar [y] += ((1.0 - nn->theta) * p_ldelta [y]);
          p_ldelta [y] = 0.0;
        }
    }

  /* Next add deltas of weights between hidden layer2 and hidden
     layer 1 */
  size = nn->num_hidden1;
  for (x = nn->num_hidden2 - 1; x >= 0; --x)
    {
      p_ldelta = &nn->hidden1_learning_delta [x * size];
      p_ldelta_bar = &nn->hidden1_learning_delta_bar [x * size];
      weight = &nn->hidden1_weights [x * size];
      p_lrate = &nn->hidden1_learning_rate [x * size];
      p_sum_delta = &nn->hidden1_weights_sum_delta [x * size];

      for (y = nn->num_hidden1 - 1; y >= 0; --y)
        {
          rate = p_ldelta [y] * p_ldelta_bar [y];
          if ( rate < 0.0 )
            {
              p_lrate [y] -= (nn->phi * p_lrate [y]);
            }
          else if ( rate > 0.0 )
            {
              p_lrate [y] += nn->K;
            }

          weight [y] += (p_lrate [y] * p_sum_delta [y]);
          p_sum_delta [y] = 0.0;
          p_ldelta_bar [y] *= nn->theta;
          p_ldelta_bar [y] += ((1.0 - nn->theta) * p_ldelta [y]);
          p_ldelta [y] = 0.0;
        }
    }

  /* Finally add deltas of weights between hidden layer 1 and input
     layer */
  size = nn->num_inputs;
  for (x = nn->num_hidden1 - 1; x >= 0; --x)
    {
      p_ldelta = &nn->input_learning_delta [x * size];
      p_ldelta_bar = &nn->input_learning_delta_bar [x * size];
      weight = &nn->input_weights [x * size];
      p_lrate = &nn->input_learning_rate [x * size];
      p_sum_delta = &nn->input_weights_sum_delta [x * size];

      for (y = nn->num_inputs - 1; y >= 0; --y)
        {
          rate = p_ldelta [y] * p_ldelta_bar [y];
          if ( rate < 0.0 )
            {
              p_lrate [y] -= (nn->phi * p_lrate [y]);
            }
          else if ( rate > 0.0 )
            {
              p_lrate [y] += nn->K;
            }

          weight [y] += (p_lrate [y] * p_sum_delta [y]);
          p_sum_delta [y] = 0.0;
          p_ldelta_bar [y] *= nn->theta;
          p_ldelta_bar [y] += ((1.0 - nn->theta) * p_ldelta [y]);
          p_ldelta [y] = 0.0;
        }
    }

  /* Now all deltas have been added into their appropriate neuron
     connection. */
  nn->training_examples += nn->examples_since_update;
  nn->examples_since_update = 0;

}


int Neural_net_calc_forward_test (Neural_net *nn, double input [],
                                  double desired_output [],
                                  int print_it, double correct_eps,
                                  double good_eps)
{
  int x,y,wrong,good;

  wrong = 0;
  good = 0;

  /* Go backward for faster processing */
  /* Calculate hidden layer 1's activation */
  for (x = nn->num_hidden1 - 1; x >= 0;  --x)
    {
      nn->hidden1_act [x] = 0.0;
      for (y = nn->num_inputs - 1; y >= 0; --y)
        {
          nn->hidden1_act [x] += (input [y] *
                              nn->input_weights [x * nn->num_inputs + y]);
        }
      nn->hidden1_act [x] = S(nn->hidden1_act [x]);
    }

  /* Calculate hidden layer 2's activation */
  for (x = nn->num_hidden2 - 1; x >= 0; --x)
    {
      nn->hidden2_act [x] = 0.0;
      for (y = nn->num_hidden1 - 1; y >= 0; --y)
        {
          nn->hidden2_act [x] += (nn->hidden1_act [y] *
                                 nn->hidden1_weights [x*nn->num_hidden1 + y]);
        }
      nn->hidden2_act [x] = S(nn->hidden2_act [x]);
    }

  /* Calculate output layer's activation */
  for (x = nn->num_outputs - 1; x >= 0; --x)
    {
      nn->output_act [x] = 0.0;
      for (y = nn->num_hidden2 - 1; y >= 0; --y)
        {
          nn->output_act [x] += nn->hidden2_act [y] *
                                nn->hidden2_weights [x * nn->num_hidden2 + y];
        }
      nn->output_act [x] = S(nn->output_act [x]);

      if ( fabs (nn->output_act [x] - desired_output [x]) > good_eps )
          wrong = 1;
      else if ( fabs (nn->output_act [x] - desired_output [x]) > correct_eps )
          good = 1;
    }

  if ( print_it )
    {
      for (x = 0; x < nn->num_outputs; ++x)
        {
          printf ("%6.4f ",nn->output_act [x]);
        }
    }

  if ( wrong )
      return (WRONG);
  else if ( good )
      return (GOOD);
  else
      return (CORRECT);
}


void Neural_net_kick_weights (Neural_net *nn, double range)
{
  int    x;
  double rand_max;
  double variation;

  rand_max = RAND_MAX;
  /* Add from -range to +range to all weights randomly */
  for (x = 0; x < (nn->num_hidden1 * nn->num_inputs); ++x)
    {
      variation = rand () / rand_max * range;
      if ( rand () < (RAND_MAX / 2) )
          variation *= -1.0;
      nn->input_weights [x] += variation;
    }

  for (x = 0; x < (nn->num_hidden2 * nn->num_hidden1); ++x)
    {
      variation = rand () / rand_max * range;
      if ( rand () < (RAND_MAX / 2) )
          variation *= -1.0;
      nn->hidden1_weights [x] += variation;
    }

  for (x = 0; x < (nn->num_outputs * nn->num_hidden2); ++x)
    {
      variation = rand () / rand_max * range;
      if ( rand () < (RAND_MAX / 2) )
          variation *= -1.0;
      nn->hidden2_weights [x] += variation;
    }

}