save on exit and output to standard error

- closing the program will now calculate first the fittest bird then it will save the neural network of that bird.

- converted all standard output 'cout' to output on standard error 'cerr' instead.

Author: mrdcvlsc

src/ffnn.cpp

@@ -4,8 +4,8 @@
 
 #include "ffnn.hpp"
 
-std::uniform_int_distribution<size_t> FFNN::random_chance(0U, 100U);
-std::uniform_real_distribution<float> FFNN::new_random_weight(-1.f, 1.f);
+std::uniform_real_distribution<float> FFNN::random_chance(0.f, 1.f);
+std::uniform_real_distribution<float> FFNN::random_weight(-1.f, 1.f);
 
 FFNN::FFNN() : input_layer(), w1(), hidden_layer(), w2(), output_layer()
 {
@@ -16,13 +16,13 @@ void FFNN::randomize_network()
 {
     for (size_t j = 0; j < INPUTS; ++j) {
         for (size_t i = 0; i < HIDDEN; ++i) {
-            w1(i, j) = new_random_weight(rand_engine);
+            w1(i, j) = random_weight(rand_engine);
         }
     }
 
     for (size_t j = 0; j < HIDDEN; ++j) {
         for (size_t i = 0; i < OUTPUT; ++i) {
-            w2(i, j) = new_random_weight(rand_engine);
+            w2(i, j) = random_weight(rand_engine);
         }
     }
 }
@@ -60,18 +60,18 @@ void FFNN::mutate()
     // mutate weight 1
     for (size_t j = 0; j < INPUTS; ++j) {
         for (size_t i = 0; i < HIDDEN; ++i) {
-            size_t chance = random_chance(rand_engine);
+            float chance = random_chance(rand_engine);
             if (chance <= MUTATION_CHANCE_THRESHOLD) {
-                w1(i, j) = new_random_weight(rand_engine);
+                w1(i, j) = random_weight(rand_engine);
             }
         }
     }
 
     for (size_t j = 0; j < HIDDEN; ++j) {
         for (size_t i = 0; i < OUTPUT; ++i) {
-            size_t chance = random_chance(rand_engine);
+            float chance = random_chance(rand_engine);
             if (chance <= MUTATION_CHANCE_THRESHOLD) {
-                w2(i, j) = new_random_weight(rand_engine);
+                w2(i, j) = random_weight(rand_engine);
             }
         }
     }
@@ -82,7 +82,7 @@ void FFNN::combine(FFNN const &nn1, FFNN const &nn2)
     // mutate weight 1
     for (size_t j = 0; j < INPUTS; ++j) {
         for (size_t i = 0; i < HIDDEN; ++i) {
-            size_t chance = random_chance(rand_engine);
+            float chance = random_chance(rand_engine);
             if (chance <= WEIGHT_SELECTION_CHANCE_THRESHOLD) {
                 w1(i, j) = nn1.w1(i, j);
             } else {
@@ -93,7 +93,7 @@ void FFNN::combine(FFNN const &nn1, FFNN const &nn2)
 
     for (size_t j = 0; j < HIDDEN; ++j) {
         for (size_t i = 0; i < OUTPUT; ++i) {
-            size_t chance = random_chance(rand_engine);
+            float chance = random_chance(rand_engine);
             if (chance <= WEIGHT_SELECTION_CHANCE_THRESHOLD) {
                 w2(i, j) = nn1.w2(i, j);
             } else {
@@ -111,7 +111,7 @@ bool FFNN::save_network(std::string const &output_filename)
     nn_file.write(reinterpret_cast<char *>(w2.data()), w2.size() * sizeof(float));
     nn_file.close();
 
-    std::cout << "------------ saved  ------------\n";
+    std::cout << "------------ saved ------------\n";
     std::cout << "w1 = \n\n" << w1 << "\n\nw2 = \n\n" << w2 << "\n\n";
 
     return nn_file.good();
@@ -125,12 +125,12 @@ bool FFNN::load_network(std::string const &filename)
         return false;
     }
 
+    std::cout << "------------ loaded ------------\n";
+    std::cout << "w1 = \n\n" << w1 << "\n\nw2 = \n\n" << w2 << "\n\n";
+
     nn_file.read(reinterpret_cast<char *>(w1.data()), w1.size() * sizeof(float));
     nn_file.read(reinterpret_cast<char *>(w2.data()), w2.size() * sizeof(float));
     nn_file.close();
 
-    std::cout << "------------ loaded ------------\n";
-    std::cout << "w1 = \n\n" << w1 << "\n\nw2 = \n\n" << w2 << "\n\n";
-
     return nn_file.good();
 }

src/ffnn.hpp

@@ -23,16 +23,16 @@ struct FFNN
     static constexpr size_t OUTPUT = 1;
 
     /// \brief 30% chance to mutate a weight.
-    static constexpr size_t MUTATION_CHANCE_THRESHOLD = 30;
+    static constexpr float MUTATION_CHANCE_THRESHOLD = 0.3f;
 
-    /// \brief 50% chance to choose from parentA's weight.
-    static constexpr size_t WEIGHT_SELECTION_CHANCE_THRESHOLD = 50;
+    /// \brief 50%/50% chance to choose from parent A's or parent B's weight.
+    static constexpr float WEIGHT_SELECTION_CHANCE_THRESHOLD = 0.5f;
 
     /// \brief generates number from 1 to 100.
-    static std::uniform_int_distribution<size_t> random_chance;
+    static std::uniform_real_distribution<float> random_chance;
 
     /// \brief generates weight from -0.5 to 0.5.
-    static std::uniform_real_distribution<float> new_random_weight;
+    static std::uniform_real_distribution<float> random_weight;
 
     Eigen::Matrix<float, INPUTS, 1>      input_layer;
     Eigen::Matrix<float, HIDDEN, INPUTS> w1;

src/main.cc

@@ -61,7 +61,7 @@ int main(int argc, char *args[])
     bool        capture_frames = false;
     if (argc == 2) {
         if (std::string(args[1]) == "capture") {
-            std::cout << "frame capture flag enabled\n";
+            std::cerr << "frame capture flag enabled\n";
             capture_frames = true;
             capture_texture.create(WINDOW_WIDTH, WINDOW_HEIGHT);
         }
@@ -78,9 +78,9 @@ int main(int argc, char *args[])
     GeneticAlgorithm genetic_algorithm;
 
     if (!birds.collection[0].neural_net.load_network("fittest.nn")) {
-        std::cout << "error loading the fittest network\n";
+        std::cerr << "error loading the fittest network\n";
     } else {
-        std::cout << "the previous fittests network is loaded\n";
+        std::cerr << "the previous fittests network is loaded\n";
     }
 
     while (window.isOpen()) {
@@ -90,6 +90,14 @@ int main(int argc, char *args[])
         sf::Event event;
         while (window.pollEvent(event)) {
             if (event.type == sf::Event::Closed) {
+                genetic_algorithm.rank_fitness(birds);
+
+                if (birds.collection[0].neural_net.save_network("fittest.nn")) {
+                    std::cerr << "fittest network saved\n";
+                } else {
+                    std::cerr << "failed to save the fittest network\n";
+                }
+
                 window.close();
             }
 
@@ -175,9 +183,9 @@ int main(int argc, char *args[])
             genetic_algorithm.rank_fitness(birds);
 
             if (birds.collection[0].neural_net.save_network("fittest.nn")) {
-                std::cout << "fittest network saved\n";
+                std::cerr << "fittest network saved\n";
             } else {
-                std::cout << "failed to save the fittest network\n";
+                std::cerr << "failed to save the fittest network\n";
             }
 
             genetic_algorithm.apply_mutations(birds);