ValueListPtr NeuralNet::run(const ValueListPtr input) const
{
LayerList::const_iterator it = m_layers.begin();
ValueListPtr vl = layer_output(*it, input);
it++;
while(it != m_layers.end())
{
vl = layer_output(*it, vl);
it++;
}
return vl;
}
std::vector<double> SimpleRecurrentNeuralNetwork::Evaluate(const std::vector<double> &input) {
std::vector<double> layer_hidden(dimension_hidden_layer_, 0.0);
std::vector<double> layer_output(dimension_output_layer_, 0.0);
for (unsigned int i = 0; i < dimension_hidden_layer_; ++i) {
layer_hidden[i] = context_[i];
if (input_layer_enable_bias_ == true) {
layer_hidden[i] += layer_weights_input_hidden_[i * (dimension_input_layer_ + input_layer_enable_bias_)];
}
for (unsigned int j = 0; j < dimension_input_layer_; ++j) {
const unsigned int ij = i * (dimension_input_layer_ + input_layer_enable_bias_) + 1 + j;
layer_hidden[i] += input[j] * layer_weights_input_hidden_[ij];
}
// Activation function
switch (hidden_layer_activation_) {
case ActivationFunctionType::Linear:
// do nothing
break;
case ActivationFunctionType::Sigmoid:
layer_hidden[i] = 1.0 / (1.0 + std::exp(-layer_hidden[i]));
break;
}
context_[i] = layer_hidden[i];
}
for (unsigned int i = 0; i < dimension_output_layer_; ++i) {
if (hidden_layer_enable_bias_ == true) {
layer_output[i] += layer_weights_hidden_output_[i * (dimension_hidden_layer_ + hidden_layer_enable_bias_)];
}
for (unsigned int j = 0; j < dimension_hidden_layer_; ++j) {
const unsigned int ij = i * (dimension_hidden_layer_ + hidden_layer_enable_bias_) + 1 + j;
layer_output[i] += layer_hidden[j] * layer_weights_hidden_output_[ij];
}
// Activation function
switch (output_layer_activation_) {
case ActivationFunctionType::Linear:
// do nothing
break;
case ActivationFunctionType::Sigmoid:
layer_output[i] = 1.0 / (1.0 + std::exp(-layer_output[i]));
break;
}
}
return layer_output;
}
