Converted max error to avg error in training. And bug fixes

Author: izzat5233

nn/network.h

@@ -100,12 +100,12 @@ class nn::Network {
 
     /**
      * Trains the neural network on the given set of input-output pairs.
-     * Uses the given lossFunction to calculate the worst error result of all iterations.
+     * Uses the given lossFunction to calculate the average error result of all iterations.
      * A call to this method represents a single epoch (full iteration on all data pairs).
      *
      * @param data Vector of input-output pairs for training.
      * @param lossFunction The loss function used to calculate the errors.
-     * @return The worst error result of all iterations, calculated by the lossFunction.
+     * @return The average error result of all iterations, calculated by the lossFunction.
      */
     double train(const vpvd_t &data, loss::function_t lossFunction);
 

nn/src/network.cpp

@@ -82,11 +82,11 @@ vd_t Network::train(const vd_t &input, const vd_t &output) {
 }
 
 double Network::train(const vpvd_t &data, loss::function_t lossFunction) {
-    double worst = -1;
+    double avg = 0;
     for (const auto &[input, output]: data) {
         vd_t res = train(input, output);
         double error = lossFunction(output, res);
-        worst = std::max(worst, error);
+        avg += error / data.size();
     }
-    return worst;
+    return avg;
 }

wasm/io_utility.h

@@ -20,22 +20,20 @@ void processCsvData(UPLOAD_HANDLER_PARAMETERS) {
     std::string line;
 
     while (std::getline(ss, line)) {
-        std::vector<double> row;
+        nn::vd_t row;
         std::stringstream lineStream(line);
         std::string cell;
 
-        bool add = true;
         while (std::getline(lineStream, cell, ',')) {
             try {
                 double value = std::stod(cell);
                 row.push_back(value);
             } catch (const std::invalid_argument &ia) {
                 EM_ASM_ARGS({ console.error("Invalid argument: " + $0) }, ia.what());
-                add = false;
             }
         }
 
-        if (add && !row.empty()) {
+        if (!row.empty()) {
             data->push_back(row);
         }
     }

wasm/network_interface.h

@@ -15,9 +15,9 @@
 class NetworkController {
 private:
     nn::vi_t dimensions;
-    nn::act::Function actFunction{};
-    nn::loss::function_t lossFunction{};
-    double alpha{};
+    nn::act::Function actFunction = nn::act::relu;
+    nn::loss::function_t lossFunction = nn::loss::sse;
+    double alpha = 0.1;
 
     nn::vvd_t *inputTrainingData = new nn::vvd_t();
     nn::vvd_t *outputTrainingData = new nn::vvd_t();
@@ -79,22 +79,18 @@ class NetworkController {
 
     void promptInputTrainingData() {
         emscripten_browser_file::upload(".csv,.txt", processCsvData, inputTrainingData);
-        if (inputTrainingData->size() == outputTrainingData->size()) { prepareTrainingData(); }
     }
 
     void promptOutputTrainingData() {
         emscripten_browser_file::upload(".csv,.txt", processCsvData, outputTrainingData);
-        if (inputTrainingData->size() == outputTrainingData->size()) { prepareTrainingData(); }
     }
 
     void promptInputTestingData() {
         emscripten_browser_file::upload(".csv,.txt", processCsvData, inputTestingData);
-        if (inputTestingData->size() == outputTestingData->size()) { prepareTestingData(); }
     }
 
     void promptOutputTestingData() {
         emscripten_browser_file::upload(".csv,.txt", processCsvData, outputTestingData);
-        if (inputTestingData->size() == outputTestingData->size()) { prepareTestingData(); }
     }
 
     void prepareTrainingData() {
@@ -124,19 +120,20 @@ class NetworkController {
 
     [[nodiscard]] nn::vvd_t predictTestingOutputs() const {
         nn::vvd_t res;
-        for (auto &in: *inputTestingData) {
+        for (const auto &in: *inputTestingData) {
             res.push_back(network->predict(in));
         }
         return res;
     }
 
     [[nodiscard]] double testingDataError() const {
-        double error = LONG_MAX;
+        double avgError = 0;
         nn::vvd_t output = predictTestingOutputs();
         for (std::size_t i = 0; i < output.size(); ++i) {
-            error = std::min(error, lossFunction(output[i], (*outputTestingData)[i]));
+            double error = lossFunction(output[i], (*testingData)[i].second);
+            avgError += error / output.size();
         }
-        return error;
+        return avgError;
     }
 };
 
@@ -146,8 +143,7 @@ EMSCRIPTEN_BINDINGS(my_module) {
     register_vector<int>("VecInt");
     register_vector<nn::ui_t>("VecUInt");
     register_vector<double>("VecNum");
-    register_vector<nn::vvd_t>("VecVecNum");
-
+    register_vector<nn::vd_t>("VecVecNum");
 
     class_<NetworkController>("Network")
             .constructor<>()