void BasicBrain::makeNN() {
assert(layout_.numLayers_ >= 3);
assert(layout_.numLayers_ < 20);
unsigned int* layerArray = new unsigned int[layout_.numLayers_];
layerArray[0] = layout_.numInputs_;
for(size_t i = 1; i < (layout_.numLayers_ - 1); i++) {
layerArray[i] = layout_.neuronsPerHidden;
}
layerArray[layout_.numLayers_ - 1] = layout_.numOutputs;
nn_ = fann_create_standard_array(layout_.numLayers_, layerArray);
fann_set_activation_function_hidden(nn_, FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output(nn_, FANN_SIGMOID_SYMMETRIC);
inputs_ = new fann_type[layout_.numInputs_];
reset();
delete[] layerArray;
#ifdef _CHECK_BRAIN_ALLOC
size_t id = ++nnAllocCnt_;
nnAllocs_[nn_] = id;
std::cerr << "alloc: " << id << std::endl;
#endif
}
FANN_EXTERNAL struct fann *FANN_API fann_create_standard(unsigned int num_layers, ...)
{
struct fann *ann;
va_list layer_sizes;
int i;
unsigned int *layers = (unsigned int *) fann_calloc(num_layers, sizeof(unsigned int));
if(layers == NULL)
{
fann_error(NULL, FANN_E_CANT_ALLOCATE_MEM);
return NULL;
}
va_start(layer_sizes, num_layers);
for(i = 0; i < (int) num_layers; i++)
{
layers[i] = va_arg(layer_sizes, unsigned int);
}
va_end(layer_sizes);
ann = fann_create_standard_array(num_layers, layers);
fann_free(layers);
return ann;
}
int main(int argc, char *argv[])
{
unsigned int layers[3] = {38, 17, 9};
struct fann *ann = fann_create_standard_array(3, layers);
fann_set_activation_function_hidden(ann, FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output(ann, FANN_LINEAR);
fann_train_on_file(ann, "scotland.data", 100000, 100, 0.00001);
fann_save(ann, "scotland_test.net");
fann_destroy(ann);
return 0;
}
/***
* @function rspamd_fann.create(nlayers, [layer1, ... layern])
* Creates new neural network with `nlayers` that contains `layer1`...`layern`
* neurons in each layer
* @param {number} nlayers number of layers
* @param {number} layerI number of neurons in each layer
* @return {fann} fann object
*/
static gint
lua_fann_create (lua_State *L)
{
#ifndef WITH_FANN
return 0;
#else
struct fann *f, **pfann;
guint nlayers, *layers, i;
nlayers = luaL_checknumber (L, 1);
if (nlayers > 0) {
layers = g_malloc (nlayers * sizeof (layers[0]));
if (lua_type (L, 2) == LUA_TNUMBER) {
for (i = 0; i < nlayers; i ++) {
layers[i] = luaL_checknumber (L, i + 2);
}
}
else if (lua_type (L, 2) == LUA_TTABLE) {
for (i = 0; i < nlayers; i ++) {
lua_rawgeti (L, 2, i + 1);
layers[i] = luaL_checknumber (L, -1);
lua_pop (L, 1);
}
}
f = fann_create_standard_array (nlayers, layers);
fann_set_activation_function_hidden (f, FANN_SIGMOID_SYMMETRIC);
fann_set_activation_function_output (f, FANN_SIGMOID_SYMMETRIC);
fann_set_training_algorithm (f, FANN_TRAIN_INCREMENTAL);
fann_randomize_weights (f, 0, 1);
if (f != NULL) {
pfann = lua_newuserdata (L, sizeof (gpointer));
*pfann = f;
rspamd_lua_setclass (L, "rspamd{fann}", -1);
}
else {
lua_pushnil (L);
}
g_free (layers);
}
else {
lua_pushnil (L);
}
return 1;
#endif
}
bool ViFann::setStructure(const Type &type, const QList<int> &neurons, const qreal &connectionRate)
{
#ifdef GPU
if(type != Standard)
{
LOG("The GPU version of FANN currently doesn't support shortcut, sparse or cascade networks.", QtFatalMsg);
exit(-1);
}
#endif
clear();
mType = type;
mInputCount = neurons.first();
mOutputCount = neurons.last();
mNeurons.clear();
for(mI = 0; mI < neurons.size(); ++mI)
{
if(neurons[mI] != 0) mNeurons.append(neurons[mI]);
}
if(mInput == NULL) delete [] mInput;
if(mOutput == NULL) delete [] mOutput;
mInput = new float[mInputCount];
mOutput = new float[mOutputCount];
unsigned int layers = mNeurons.size();
unsigned int layerNeurons[layers];
for(mI = 0; mI < layers; ++mI) layerNeurons[mI] = mNeurons[mI];
if(type == Standard) mNetwork = fann_create_standard_array(layers, layerNeurons);
#ifndef GPU
else if(type == Sparse)
{
mNetwork = fann_create_sparse_array(connectionRate, layers, layerNeurons);
mConnectionRate = connectionRate;
}
else if(type == Shortcut) mNetwork = fann_create_shortcut_array(layers, layerNeurons);
#endif
else return false;
fann_set_train_stop_function(mNetwork, FANN_STOPFUNC_MSE);
if(ENABLE_CALLBACK)
{
fann_set_callback(mNetwork, &ViFann::trainCallback);
mMseTotal.clear();
mMseCount = 0;
}
return true;
}
/*! fann.create_standard(num_layers, neurons_1, neurons_2, ..., neurons_n)
*# Creates a neural network with {{num_layers}}.\n
*# The i'th layer will have {{neurons_i}} neurons (the function must thus have
*# {{num_layers+1}} parameters in total).
*x ann = fann.create_standard(3, 2, 3, 1)
*-
*/
static int ann_create_standard(lua_State *L)
{
struct fann **ann;
int num_layers, i;
unsigned int *layers;
luaL_argcheck(L, lua_isinteger(L,1), 1, "First argument to fann.create_standard() must be an integer");
num_layers = lua_tointeger(L, 1);
#ifdef FANN_VERBOSE
printf("Creating neural net, %d layers\n", num_layers);
#endif
if(num_layers < 1)
luaL_error(L, "Neural network must have at least one layer");
if(lua_gettop(L) < num_layers + 1)
luaL_error(L, "Neural net has %d layers, so fann.open() must have %d parameters", num_layers, num_layers + 1);
layers = lua_newuserdata(L, num_layers*(sizeof *layers));
for(i = 0; i < num_layers; i++)
{
int n = luaL_checkinteger(L, i + 2);
if(n < 1)
{
luaL_error(L, "Layer %d must have at least 1 neuron", i);
}
#ifdef FANN_VERBOSE
printf("Layer %d to have %d neurons\n", i, n);
#endif
layers[i] = n;
}
ann = lua_newuserdata(L, sizeof *ann);
luaL_getmetatable(L, FANN_METATABLE);
lua_setmetatable(L, -2);
*ann = fann_create_standard_array(num_layers, layers);
if(!*ann)
{
luaL_error(L, "Unable to create neural network");
}
return 1;
}
int main()
{
const float desired_error = (const float) 0.0001;
const unsigned int max_epochs = 350;
const unsigned int epochs_between_reports = 25;
struct fann *ann;
struct fann_train_data *train_data;
unsigned int i = 0;
printf("Creating network.\n");
train_data = fann_read_train_from_file("ann_training_data");
// Using incremental training -> shuffle training data
fann_shuffle_train_data(train_data);
// ann = fann_create_standard(num_layers,
// train_data->num_input, num_neurons_hidden, train_data->num_output);
//ann = fann_create_standard(500, 2, 50, 50, 21);
unsigned int layers[4] = {150, 70, 30, 22};
ann = fann_create_standard_array(4, layers);
printf("Training network.\n");
fann_set_activation_function_hidden(ann, FANN_SIGMOID_SYMMETRIC_STEPWISE);
fann_set_activation_function_output(ann, FANN_LINEAR_PIECE); //FANN_SIGMOID_STEPWISE);
fann_set_training_algorithm(ann, FANN_TRAIN_INCREMENTAL);
fann_train_on_data(ann, train_data, max_epochs, epochs_between_reports, desired_error);
printf("Saving network.\n");
fann_save(ann, "mymoves_gestures.net");
printf("Cleaning up.\n");
fann_destroy_train(train_data);
fann_destroy(ann);
return 0;
}
int main()
{
const float desired_error = (const float) 0.01;
const unsigned int max_epochs = 50000;
const unsigned int epochs_between_reports = 10;
unsigned int layers []= {6, 200, 200, 4};
struct fann * ann = fann_create_standard_array(4, layers);
fann_set_activation_function_layer(ann, FANN_SIGMOID_SYMMETRIC, 1);
fann_set_activation_function_layer(ann, FANN_SIGMOID_SYMMETRIC, 2);
fann_set_activation_function_layer(ann, FANN_SIGMOID_SYMMETRIC, 3);
fann_train_on_file(ann, "../training_file", max_epochs,
epochs_between_reports, desired_error);
fann_save(ann, "wii.net");
fann_destroy(ann);
return 0;
}
/***
* @function rspamd_fann.create(nlayers, [layer1, ... layern])
* Creates new neural network with `nlayers` that contains `layer1`...`layern`
* neurons in each layer
* @param {number} nlayers number of layers
* @param {number} layerI number of neurons in each layer
* @return {fann} fann object
*/
static gint
lua_fann_create (lua_State *L)
{
#ifndef WITH_FANN
return 0;
#else
struct fann *f, **pfann;
guint nlayers, *layers, i;
nlayers = luaL_checknumber (L, 1);
if (nlayers > 0) {
layers = g_malloc (nlayers * sizeof (layers[0]));
for (i = 0; i < nlayers; i ++) {
layers[i] = luaL_checknumber (L, i + 2);
}
f = fann_create_standard_array (nlayers, layers);
if (f != NULL) {
pfann = lua_newuserdata (L, sizeof (gpointer));
*pfann = f;
rspamd_lua_setclass (L, "rspamd{fann}", -1);
}
else {
lua_pushnil (L);
}
}
else {
lua_pushnil (L);
}
return 1;
#endif
}
/* Creates a feedforward, layered net which is an "unrolled" recurrent network.
For example, the recurrent net:
A <-> B <-> C<- (where C autosynapses)
Becomes (unrolled two time steps):
A B C input layer
\/ \/|
A B C hidden layer I
\/ \/|
A B C output layer
*/
FANN_EXTERNAL struct fann *FANN_API fann_create_unrolled_recurrent(
unsigned int num_neurons, fann_type *weights, unsigned int time_steps)
{
struct fann *ann = NULL;
unsigned int *layers = NULL;
unsigned int num_layers = time_steps + 1;
unsigned int layern = 0;
struct fann_layer *curr_layer = NULL;
struct fann_neuron *curr_neuron = NULL;
fann_type *curr_weights = weights;
/*************************************
CREATE THE FEEDFORWARD STRUCTURE
*************************************/
/* Allocate number of neurons per layer array */
layers = (unsigned int *)calloc(num_layers, sizeof(unsigned int));
if (layers == NULL)
{
return NULL;
}
/* Populate each layer with the number of neurons */
for (layern=0; layern < num_layers; layern++)
{
layers[layern] = num_neurons;
}
/* Create the feedforward network */
ann = fann_create_standard_array(num_layers, layers);
fann_safe_free(layers);
/*printf("REQUESTED: LAYERS=%d, NEURONS/LAYER=%d\n", num_layers, num_neurons);
printf("NUM LAYERS: %d\n", ann->last_layer - ann->first_layer);
printf("IN: %d, NEURONS: %d, OUTPUT: %d\n",
ann->num_input, ann->num_neurons, ann->num_output);*/
/*************************************
SET THE FEEDFORWARD WEIGHTS
*************************************/
/* Visit each layer */
for (curr_layer = ann->first_layer;
curr_layer != ann->last_layer;
curr_layer++)
{
/* The weights are the same for each feedforward layer! */
curr_weights = weights;
/* Copy the weight matrix into the neurons,
one row per neuron */
for (curr_neuron = curr_layer->first_neuron;
curr_neuron != curr_layer->last_neuron;
curr_neuron++)
{
memcpy(curr_neuron->weights, curr_weights, num_neurons * num_neurons * sizeof(fann_type));
curr_weights += num_neurons;
}
}
return ann;
}
/*
arguments (all required):
- data filename
- topology, as number of neurons per layer separated by dashes
- epochs (integer)
- learning rate (0.0-1.0 float)
- output filename
*/
int main(int argc, char **argv)
{
// Argument 1: data filename.
const char *datafn = argv[1];
// Argument 2: topology.
unsigned int layer_sizes[MAX_LAYERS];
unsigned int num_layers = 0;
char *token = strtok(argv[2], "-");
while (token != NULL) {
layer_sizes[num_layers] = atoi(token);
++num_layers;
token = strtok(NULL, "-");
}
// Argument 3: epoch count.
unsigned int max_epochs = atoi(argv[3]);
// Argument 4: learning rate.
float learning_rate = atof(argv[4]);
// Argument 5: output filename.
const char *outfn = argv[5];
struct fann *ann;
ann = fann_create_standard_array(num_layers, layer_sizes);
// Misc parameters.
fann_set_training_algorithm(ann, FANN_TRAIN_RPROP);
fann_set_activation_steepness_hidden(ann, 0.5);
fann_set_activation_steepness_output(ann, 0.5);
fann_set_activation_function_hidden(ann, FANN_SIGMOID);
fann_set_activation_function_output(ann, FANN_SIGMOID);
//fann_set_train_stop_function(ann, FANN_STOPFUNC_BIT);
//fann_set_bit_fail_limit(ann, 0.01f);
struct fann_train_data *data;
data = fann_read_train_from_file(datafn);
fann_init_weights(ann, data);
fann_set_learning_rate(ann, learning_rate);
fann_train_on_data(
ann,
data,
max_epochs,
10, // epochs between reports
DESIRED_ERROR
);
printf("Testing network. %f\n", fann_test_data(ann, data));
fann_type *calc_out;
for(unsigned int i = 0; i < fann_length_train_data(data); ++i)
{
calc_out = fann_run(ann, data->input[i]);
}
printf("RMSE = %f\n", sqrt(fann_get_MSE(ann)));
fann_save(ann, outfn);
fann_destroy_train(data);
fann_destroy(ann);
return 0;
}
int main(int argc, char *argv[])
{
struct fann_train_data *dadosTreino, *dadosTeste;
struct fann *ANN;
fann_type *ANN_Answers;
int *layers, i, j, aux;
chromosome chromo;
float erro = 0.0;
checkArgs(argc, argv);
buildChromosome(argv, &chromo);
checkDatasetFiles();
dadosTreino = fann_read_train_from_file(nomeArqTreino);
layers = (int *) calloc(2+chromo.qntCamadasOcultas, sizeof(int));
layers[0] = qntNeuroniosEntrada;
layers[2+chromo.qntCamadasOcultas-1] = qntNeuroniosSaida;
aux = chromo.neurOcultos;
for (i=1; i < 2+chromo.qntCamadasOcultas-1 ; i++)
{
layers[i] = aux;
aux = aux/2;
}
// CRIANDO A RNA:
ANN = fann_create_standard_array(2+chromo.qntCamadasOcultas, layers);
// TREINO
fann_set_learning_rate(ANN, chromo.learnRate);
fann_set_learning_momentum(ANN, chromo.moment);
fann_set_activation_function_hidden( ANN, chromo.fcOculta );
fann_set_activation_function_output( ANN, chromo.fcSaida );
fann_set_training_algorithm(ANN, chromo.algAprend );
if (fann_get_training_algorithm(ANN) == FANN_TRAIN_QUICKPROP)
fann_set_quickprop_decay(ANN, chromo.decay);
// Em python, o treino ficava entre um try.
// Se desse erro, escrevia "Resultado: 999.0" e exit
fann_train_on_data(ANN, dadosTreino, chromo.epocasTreino, 50, desiredError);
fann_destroy_train(dadosTreino);
// TESTES:
dadosTeste = fann_read_train_from_file( nomeArqValidacao);
// Em python, o teste também ficava entre um try.
// Se desse erro, escrevia "Resultado: 999.0" e exit
for(i = 0; i < fann_length_train_data(dadosTeste); i++)
{
ANN_Answers = fann_run(ANN, dadosTeste->input[i]);
if (ANN_Answers == NULL)
{
printf("Resultado: 999.0\n");
exit(2);
}
for (j=0; j < qntNeuroniosSaida; j++)
erro += (float) powf(fann_abs(ANN_Answers[j] - dadosTeste->output[i][j]), 2);
}
printf("Resultado: %f\n", erro/(fann_length_train_data(dadosTeste)-1));
fann_destroy_train(dadosTeste);
saveANN(argc, argv, ANN);
fann_destroy(ANN);
}
/***
* @function rspamd_fann.create_full(params)
* Creates new neural network with parameters:
* - `layers` {table/numbers}: table of layers in form: {N1, N2, N3 ... Nn} where N is number of neurons in a layer
* - `activation_hidden` {string}: activation function type for hidden layers (`tanh` by default)
* - `activation_output` {string}: activation function type for output layer (`tanh` by default)
* - `sparsed` {float}: create sparsed ANN, where number is a coefficient for sparsing
* - `learn` {string}: learning algorithm (quickprop, rprop or incremental)
* - `randomize` {boolean}: randomize weights (true by default)
* @return {fann} fann object
*/
static gint
lua_fann_create_full (lua_State *L)
{
#ifndef WITH_FANN
return 0;
#else
struct fann *f, **pfann;
guint nlayers, *layers, i;
const gchar *activation_hidden = NULL, *activation_output, *learn_alg = NULL;
gdouble sparsed = 0.0;
gboolean randomize_ann = TRUE;
GError *err = NULL;
if (lua_type (L, 1) == LUA_TTABLE) {
lua_pushstring (L, "layers");
lua_gettable (L, 1);
if (lua_type (L, -1) != LUA_TTABLE) {
return luaL_error (L, "bad layers attribute");
}
nlayers = rspamd_lua_table_size (L, -1);
if (nlayers < 2) {
return luaL_error (L, "bad layers attribute");
}
layers = g_new0 (guint, nlayers);
for (i = 0; i < nlayers; i ++) {
lua_rawgeti (L, -1, i + 1);
layers[i] = luaL_checknumber (L, -1);
lua_pop (L, 1);
}
lua_pop (L, 1); /* Table */
if (!rspamd_lua_parse_table_arguments (L, 1, &err,
"sparsed=N;randomize=B;learn=S;activation_hidden=S;activation_output=S",
&sparsed, &randomize_ann, &learn_alg, &activation_hidden, &activation_output)) {
g_free (layers);
if (err) {
gint r;
r = luaL_error (L, "invalid arguments: %s", err->message);
g_error_free (err);
return r;
}
else {
return luaL_error (L, "invalid arguments");
}
}
if (sparsed != 0.0) {
f = fann_create_standard_array (nlayers, layers);
}
else {
f = fann_create_sparse_array (sparsed, nlayers, layers);
}
if (f != NULL) {
pfann = lua_newuserdata (L, sizeof (gpointer));
*pfann = f;
rspamd_lua_setclass (L, "rspamd{fann}", -1);
}
else {
g_free (layers);
return luaL_error (L, "cannot create fann");
}
fann_set_activation_function_hidden (f,
string_to_activation_func (activation_hidden));
fann_set_activation_function_output (f,
string_to_activation_func (activation_output));
fann_set_training_algorithm (f, string_to_learn_alg (learn_alg));
if (randomize_ann) {
fann_randomize_weights (f, 0, 1);
}
g_free (layers);
}
else {
return luaL_error (L, "bad arguments");
}
return 1;
#endif
}
int sci_fann_create(char * fname)
{
int * pi_command_addr = NULL;
int m_layers, n_layers, * pi_layers_addr = NULL;
int * pi_conn_addr = NULL;
char * Command = NULL;
double * layers = NULL, conn = 0.0;
unsigned int * ui_layers = NULL;
int res, numLayers, i;
struct fann * result_ann = NULL;
SciErr _sciErr;
if (Rhs<2)
{
Scierror(999,"%s usage: ann = %s(command,[layers ...])", fname, fname);
return 0;
}
_sciErr = getVarAddressFromPosition(pvApiCtx, 1, &pi_command_addr);
if (_sciErr.iErr)
{
printError(&_sciErr, 0);
return 0;
}
getAllocatedSingleString(pvApiCtx, pi_command_addr, &Command);
_sciErr = getVarAddressFromPosition(pvApiCtx, 2, &pi_layers_addr);
if (_sciErr.iErr)
{
printError(&_sciErr, 0);
return 0;
}
_sciErr = getMatrixOfDouble(pvApiCtx, pi_layers_addr, &m_layers, &n_layers, &layers);
if ((n_layers != 1) & (m_layers !=1))
{
Scierror(999,"%s: Layers must be a vector!",fname);
return 0;
}
numLayers = m_layers * n_layers;
ui_layers = (unsigned int *)MALLOC(numLayers*sizeof(unsigned int));
for(i=0; i<numLayers; i++) ui_layers[i] = layers[i];
if (strcmp(Command,"standard") == 0)
{
freeAllocatedSingleString(Command);
// fann_create_standard_array Just like fann_create_standard, but with an array of layer sizes instead of individual parameters.
result_ann = fann_create_standard_array(numLayers,ui_layers);
FREE(ui_layers);
if (result_ann==NULL)
{
Scierror(999,"%s: not able to create standard network\n",fname);
return 0;
}
}
if (strcmp(Command,"sparse") == 0)
{
freeAllocatedSingleString(Command);
// fann_create_sparse_array Just like fann_create_sparse, but with an array of layer sizes instead of individual parameters.
_sciErr = getVarAddressFromPosition(pvApiCtx, 3, &pi_conn_addr);
if (_sciErr.iErr)
{
printError(&_sciErr, 0);
return 0;
}
getScalarDouble(pvApiCtx, pi_conn_addr, &conn);
result_ann = fann_create_sparse_array(conn,numLayers,ui_layers);
FREE(ui_layers);
if (result_ann==NULL)
{
Scierror(999,"%s: not able to create sparse network\n",fname);
return 0;
}
}
if (strcmp(Command,"shortcut") == 0)
{
freeAllocatedSingleString(Command);
// fann_create_shortcut_array Just like fann_create_shortcut, but with an array of layer sizes instead of individual parameters.
result_ann = fann_create_shortcut_array(numLayers,ui_layers);
FREE(ui_layers);
if (result_ann==NULL)
{
Scierror(999,"%s: not able to create shortcut network\n",fname);
return 0;
}
}
//Create the struct representing this ann in scilab
res = createScilabFannStructFromCFannStruct(result_ann, Rhs + 1);
if (res==-1) return 0;
LhsVar(1) = Rhs + 1;
return 0;
}