static void test_backprop_neuron_init()
{
bp_neuron n;
int no_of_inputs=10;
unsigned int random_seed = 123;
printf("test_backprop_neuron_init...");
bp_neuron_init(&n, no_of_inputs, &random_seed);
bp_neuron_free(&n);
printf("Ok\n");
}
static void test_backprop_neuron_save_load()
{
bp_neuron n1, n2;
int no_of_inputs=10;
unsigned int random_seed = 123;
char filename[256];
FILE * fp;
printf("test_backprop_neuron_save_load...");
/* create neurons */
bp_neuron_init(&n1, no_of_inputs, &random_seed);
bp_neuron_init(&n2, no_of_inputs, &random_seed);
sprintf(filename,"%stemp_deep.dat",DEEPLEARN_TEMP_DIRECTORY);
/* save the first neuron */
fp = fopen(filename,"wb");
assert(fp!=0);
bp_neuron_save(fp, &n1);
fclose(fp);
/* load into the second neuron */
fp = fopen(filename,"rb");
assert(fp!=0);
bp_neuron_load(fp, &n2);
fclose(fp);
/* compare the two */
assert(bp_neuron_compare(&n1, &n2)==1);
/* free memory */
bp_neuron_free(&n1);
bp_neuron_free(&n2);
printf("Ok\n");
}
static void test_backprop_neuron_copy()
{
bp_neuron n1, n2;
int retval, no_of_inputs=10;
unsigned int random_seed = 123;
printf("test_backprop_neuron_copy...");
bp_neuron_init(&n1, no_of_inputs, &random_seed);
bp_neuron_init(&n2, no_of_inputs, &random_seed);
bp_neuron_copy(&n1, &n2);
retval = bp_neuron_compare(&n1, &n2);
if (retval != 1) {
printf("\nretval %d\n", retval);
}
assert(retval == 1);
bp_neuron_free(&n1);
bp_neuron_free(&n2);
printf("Ok\n");
}
/**
* @brief Initialise a backprop neural net
* @param net Backprop neural net object
* @param no_of_inputs The number of input units
* @param no_of_hiddens The number of units in each hidden layer
* @param hidden_layers The number of hidden layers
* @param no_of_inputs The number of output units
* @param random_seed The random number generator seed
*/
void bp_init(bp * net,
int no_of_inputs,
int no_of_hiddens,
int hidden_layers,
int no_of_outputs,
unsigned int * random_seed)
{
int i, j, l;
bp_neuron * n;
net->learningRate = 0.2f;
net->noise = 0.0f;
net->random_seed = *random_seed;
net->BPerror = DEEPLEARN_UNKNOWN_ERROR;
net->BPerrorAverage = DEEPLEARN_UNKNOWN_ERROR;
net->BPerrorTotal = DEEPLEARN_UNKNOWN_ERROR;
net->itterations = 0;
net->DropoutPercent = 20;
net->NoOfInputs = no_of_inputs;
net->inputs = (bp_neuron**)malloc(no_of_inputs*sizeof(bp_neuron*));
net->NoOfHiddens = no_of_hiddens;
net->HiddenLayers = hidden_layers;
net->hiddens =
(bp_neuron***)malloc(hidden_layers*sizeof(bp_neuron**));
for (l = 0; l < hidden_layers; l++) {
net->hiddens[l] =
(bp_neuron**)malloc(no_of_hiddens*sizeof(bp_neuron*));
}
net->NoOfOutputs = no_of_outputs;
net->outputs = (bp_neuron**)malloc(no_of_outputs*sizeof(bp_neuron*));
/* create inputs */
for (i = 0; i < net->NoOfInputs; i++) {
net->inputs[i] = (bp_neuron*)malloc(sizeof(struct bp_n));
bp_neuron_init(net->inputs[i], 1, random_seed);
}
/* create hiddens */
for (l = 0; l < hidden_layers; l++) {
for (i = 0; i < net->NoOfHiddens; i++) {
net->hiddens[l][i] =
(bp_neuron*)malloc(sizeof(bp_neuron));
n = net->hiddens[l][i];
if (l == 0) {
bp_neuron_init(n, no_of_inputs, random_seed);
/* connect to input layer */
for (j = 0; j < net->NoOfInputs; j++) {
bp_neuron_add_connection(n, j, net->inputs[j]);
}
}
else {
bp_neuron_init(n, no_of_hiddens, random_seed);
/* connect to previous hidden layer */
for (j = 0; j < net->NoOfHiddens; j++) {
bp_neuron_add_connection(n, j, net->hiddens[l-1][j]);
}
}
}
}
/* create outputs */
for (i = 0; i < net->NoOfOutputs; i++) {
net->outputs[i] = (bp_neuron*)malloc(sizeof(bp_neuron));
n = net->outputs[i];
bp_neuron_init(n, no_of_hiddens, random_seed);
for (j = 0; j < net->NoOfHiddens; j++) {
bp_neuron_add_connection(n, j,
net->hiddens[hidden_layers-1][j]);
}
}
}
