/***non recursive permutation module***/
int permutation_non_recursive(const char* s){
/*wrapper*/
int len=strlen(s);
char* tmp=(char *)malloc((len+1)*sizeof(char));
if(!tmp)
return -1;
int* address=(int *)malloc(len*sizeof(int));
if(!address)
return -1;
strcpy(tmp,s);
permutation_init(s,address,len);
/*permutate all possibilities*/
while (permutation_next(s,address,tmp,len)!=-1) {
continue;
}
free(tmp);
free(address);
printf("\n");
}
int train_network_pool(float error_limit)
{
int i, init = 0; //init is 0 for the first generation of the algorithm
char index[10],file_name[100], selected_network_file[100];
dnc* dnc_vars;
Marchand* marchand_vars;
float learning_rate, momentum, local_error, threshold = 0.001;
int selected_index, selected_source; // selected_source = 0 -> dnc, 1 -> marchand selected index x-> xth network of selected_source
int present_gen = 0;
int count_mc = 0; // count misclassification
float init_error = 1.0;
srand(time(NULL));
dnc_vars = new dnc[INIT_NET];
marchand_vars = new Marchand[INIT_NET];
nn test_network;
float present_error, previous_error;
int increase_count = 0;
present_error = previous_error = 1.0;
char file_name_to_save[100];
FILE *fp_output = fopen("graph_input.csv","w");
if(fp_output == NULL) printf("Error Opening File ... %s\n", "graph_input.csv");
fprintf(fp_output, "%d,%f\n", present_gen, 1.0*NUM_GEN);
while(present_gen < NUM_GEN)
{
if(present_gen)
//if(true)
{
float local_error = 0.0;
learning_rate = (double)(rand() % 10 + 2) / (double)(100);
momentum = (double)(rand() % 9 + 1) / (double)(10);
cout << "GEN learing rate : " << learning_rate << " momentum : " << momentum << endl;
int layer[3] = {ATTR_NUM, 1, NUM_CLASS};
test_network.create_nn(3, layer, 0, learning_rate, momentum, MAX_NODES);
printf("running ..");
test_network.load(selected_network_file);
//for( int i = 0; i < TRAIN_SIZE; i++)
for( int i = 0; i < VALIDATION_SIZE; i++)
{
test_network.calculate(validation_input[i]);
local_error += test_network.get_error(validation_output[i]);
}
printf("Total error: %f Average error: %f\n", local_error, (float)(local_error/ TRAIN_SIZE));
fprintf(fp_output, "%d,%f\n", present_gen, NUM_GEN * (float)(local_error/ TRAIN_SIZE));
if( (float)(local_error/ TRAIN_SIZE) < error_limit)
{
test_network.save((char *) "genetic_ouptut.txt", 2);
printf("The network is converged ...\n");
fclose(fp_output);
return 0;
}
else
{
test_network.save((char *) "genetic_ouptut.txt", 2);
printf("The network is SAVED ...\n");
}
present_error = (float)(local_error/ TRAIN_SIZE);
if(fabs(present_error - previous_error) < (error_limit/10))
{
printf("Network Converged ... final case\n");
test_network.save((char *) "genetic_ouptut.txt", 2);
fclose(fp_output);
return 0;
}
if(present_error > previous_error)
{
increase_count++;
sprintf(file_name_to_save, "genetic/genetic_output%d", increase_count);
test_network.save((char*) file_name_to_save);
sprintf(file_name_to_save, "");
}
else if(present_error < previous_error) increase_count = 0;
if(increase_count == MAIN_ERROR_INCREASE)
{
test_network.save((char *) "genetic_ouptut.txt", 2);
printf("Returning from the main function ... %d\n", increase_count);
fclose(fp_output);
return 0;
}
else if( (present_error > previous_error) && (present_error - previous_error) < MAIN_ERROR_CHANGE)
{
sprintf(file_name_to_save, "genetic/genetic_outputLastCase");
test_network.save((char*) file_name_to_save);
sprintf(file_name_to_save, "");
printf("Last case, Increase Count : %d\n", increase_count);
fclose(fp_output);
return 0;
}
printf("The increase count is : %d\n", increase_count);
previous_error = present_error;
}
int m;
for(i = 0; i < INIT_NET; i++)
{
sprintf(file_name, "dnc%d",i);
//puts(file_name);
//supply each network a time to stop, a random learning rate and a random permutation
learning_rate = (double)(rand() % 10 + 2) / (double)(100);
momentum = (double)(rand() % 7 + 1) / (double)(10);
cout << "learing rate : " << learning_rate << " momentum : " << momentum << endl;
dnc_vars[i].init_dnc(ATTR_NUM, NUM_CLASS, TRAIN_SIZE, EPOCH_DNC, 0.001, MAX_NODES, learning_rate, momentum, file_name, TRAIN_TIME);
dnc_vars[i].set_monotone_increament(5);
dnc_vars[i].take_validation_data((float *) validation_input, (float *)validation_output, VALIDATION_SIZE); // num of validation data as last argument
if(present_gen) dnc_vars[i].load(selected_network_file);
error[i] = dnc_vars[i].execute((float *)input, (float *)output);
strcpy(file_name,"");
}
m = i; // hold the value INIT_NET
for(i = 0; i < INIT_NET; i++)
{
strcpy(file_name,"");
sprintf(file_name, "marchand%d", i);
//puts(file_name);
if(!present_gen)
{
permutation_next();
marchand_vars[i].init_marchand(ATTR_NUM, NUM_CLASS, TRAIN_SIZE, 200, 0.05, MAX_NODES, file_name, TRAIN_TIME/3.0);
marchand_vars[i].take_validation_data((float *) validation_input, (float *)validation_output, VALIDATION_SIZE);
error[m+i] = marchand_vars[i].execute((float *)input_temp, (float *)output_temp);
}
else
{
permutation_next_another((float *)remaining_input, (float *)remaining_output, count_mc);
marchand_vars[i].init_marchand(ATTR_NUM, NUM_CLASS, count_mc, 200, 0.05, MAX_NODES, file_name, TRAIN_TIME/3.0);
marchand_vars[i].load(selected_network_file);
marchand_vars[i].take_validation_data((float *) validation_input, (float *)validation_output, VALIDATION_SIZE);
error[m+i] = marchand_vars[i].execute((float *)remaining_input, (float *)remaining_output);
}
}
//best selection
select_network(&selected_index, &selected_source);
cout << "selected index: " << selected_index << " selected source: " << selected_source << endl;
count_mc = 0;
if(selected_source == 0)
{
for(int m = 0; m < TRAIN_SIZE; m++)
{
local_error = dnc_vars[selected_index].calculate_for_single(input[m],output[m]);
if(local_error > threshold)
{
for(int k = 0; k < ATTR_NUM; k++) remaining_input[count_mc][k] = input[m][k];
for(int k = 0; k < NUM_CLASS; k++) remaining_output[count_mc][k] = output[m][k];
count_mc++;
}
}
}
if(!selected_source) sprintf(selected_network_file, "dnc%d", selected_index);
else sprintf(selected_network_file, "marchand%d", selected_index);
cout << "Generation is : " << present_gen << endl;
present_gen++;
}
delete []dnc_vars;
delete []marchand_vars;
return 0;
}