Line data    Source code

       1             : #include "NeuralNetwork.h"
       2             : #include "LayerHidden.h"
       3             : #include "LayerInput.h"
       4             : #include "LayerOutput.h"
       5             : #include "SimpleLogger.h"
       6             : #include "exception/NeuralNetworkException.h"
       7             : 
       8             : using namespace sipai;
       9             : 
      10          61 : cv::Mat NeuralNetwork::forwardPropagation(const cv::Mat &inputValues) {
      11          61 :   if (layers.front()->layerType != LayerType::LayerInput) {
      12           0 :     throw NeuralNetworkException("Invalid front layer type");
      13             :   }
      14          61 :   if (layers.back()->layerType != LayerType::LayerOutput) {
      15           0 :     throw NeuralNetworkException("Invalid back layer type");
      16             :   }
      17          61 :   ((LayerInput *)layers.front())->setInputValues(inputValues);
      18         245 :   for (auto &layer : layers) {
      19         184 :     layer->forwardPropagation();
      20             :   }
      21          61 :   return ((LayerOutput *)layers.back())->getOutputValues();
      22             : }
      23             : 
      24          42 : void NeuralNetwork::backwardPropagation(const cv::Mat &expectedValues,
      25             :                                         const float &error_min,
      26             :                                         const float &error_max) {
      27          42 :   if (layers.back()->layerType != LayerType::LayerOutput) {
      28           0 :     throw NeuralNetworkException("Invalid back layer type");
      29             :   }
      30          42 :   ((LayerOutput *)layers.back())->computeErrors(expectedValues);
      31         168 :   for (auto it = layers.rbegin(); it != layers.rend(); ++it) {
      32         126 :     (*it)->backwardPropagation(error_min, error_max);
      33             :   }
      34          42 : }
      35             : 
      36          42 : void NeuralNetwork::updateWeights(float learning_rate) {
      37         168 :   for (auto &layer : layers) {
      38         126 :     layer->updateWeights(learning_rate);
      39             :   }
      40          42 : }