runnotes

.gitignore

@@ -4,6 +4,7 @@ mnist_train.csv
 newclassify.py
 accuracyLog.txt
 old/
+logs/*
 *accuracyLog.txt
 *.out
 .vscode

Makefile

@@ -0,0 +1,12 @@
+CXX = clang++
+CXXFLAGS = -Ofast -std=c++20
+TARGET = runmain.out
+SRC = main.cpp 
+
+$(TARGET): $(SRC)
+	$(CXX) $(CXXFLAGS) -o $(TARGET) $(SRC)
+
+.PHONY: clean
+clean:
+	rm -f $(TARGET)
+

loadcsv.cpp

@@ -75,11 +75,8 @@ vector<vector<int>> one_hot_encode(const vector<int>& input, int num_classes) {
             encoded[i][input[i]] = 1;
         } else {
             cerr << "Value out of range: " << input[i] << endl;
-            // Handle the error according to your needs.
-            // Here, we just print an error message.
         }
     }
-
     return encoded;
 }
 

main.cpp

@@ -10,11 +10,11 @@ int main()
     cout << "#################################################################" << endl;
     cout << "Generating Model" << endl;
     cout << "#################################################################" << endl;
-
     // Model
     double learningRate = 0.00015666;
     string optimizerType = "adam"; // supported: sgd, sgd_momentum, adagrad, rmsprop, adam
-    Model test("cross_entropy", optimizerType, learningRate);
+    string runNote = "tanh_256_adam";
+    Model test("cross_entropy", optimizerType, learningRate, runNote);
 
     vector<CallBackFunctionType> callbacks;
     test.addLayer("tanh", make_tuple(256, 784), callbacks);
@@ -40,10 +40,10 @@ int main()
     vector<vector<int>> label_vec = one_hot_encode(labelNums, 10);
 
     // subset data for faster testing:
-    // label_vec.resize(5000);
-    // images.resize(5000);
+    label_vec.resize(5000);
+    images.resize(5000);
 
-    test.teach(label_vec, images, 100);
+    test.teach(label_vec, images, 1);
 
     cout << "train accuracy " << test.getLastAccuracy() << endl;
 

model.cpp

@@ -27,8 +27,9 @@ class Model
     vector<int> labels;
     Layer *topographyHead = NULL;
     vector<CallBackFunctionType> callBacks;
+    string runNote;
 
-    Model(const string lossFunctionID, string optimizerType, double learningRate) : optimizerType(optimizerType), learningRate(learningRate)
+    Model(const string lossFunctionID, string optimizerType, double learningRate, string runNote="undefined") : optimizerType(optimizerType), learningRate(learningRate), runNote(runNote)
     {
         auto [lossFunc, lossDeriv] = getLossFunctions(lossFunctionID);
         lossFunction = lossFunc;
@@ -164,7 +165,8 @@ class Model
             }
             cout << endl;
             ofstream myfile;
-            myfile.open("training_accuracyLog_tanh_RMS.txt", ios::app);
+            string trainingLog = "logs/" + runNote + "_training.txt";
+            myfile.open(trainingLog, ios::app);
             myfile << j << "," << float(correctPredictions) / float(totalPredictions) << "," << sumLoss / double(totalPredictions) << "\n";
             sumLoss = 0.0;
             myfile.close();
@@ -230,7 +232,8 @@ class Model
             }
             cout << endl;
             ofstream myfile;
-            myfile.open("testing_accuracyLog_tanh_RMS.txt", ios::app);
+            string testingLog = "logs/" + runNote + "_testing.txt";
+            myfile.open(testingLog, ios::app);
             myfile << j << "," << float(correctPredictions) / float(totalPredictions) << "," << sumLoss / double(totalPredictions) << "\n";
             sumLoss = 0.0;
             myfile.close();
@@ -299,3 +302,4 @@ class Model
         }
     }
 };
+

utils.cpp

@@ -1,5 +1,7 @@
 #include<vector>
 
+using namespace std;
+
 vector<double> cast_vector_to_double(const vector<int>& int_vector) {
     vector<double> double_vector;
     double_vector.reserve(int_vector.size()); // Reserve space to avoid reallocations