jacobian

commit a92faf16c14b08ff92066b0630f71474b9fbe3a1
parent 6338610c6f712a734ba91b882cb214ffcbf3035d
Author: David Freifeld <freifeld.david@gmail.com>
Date:   Wed, 15 Jul 2020 12:49:27 -0700

Large bugfix + much better accuracy

Diffstat:
M
example.cpp
 | 
3
+--
M
src/bpnn.cpp
 | 
23
+++++++++--------------

2 files changed, 10 insertions(+), 16 deletions(-)
diff --git a/example.cpp b/example.cpp
@@ -14,7 +14,7 @@
 double bench(int batch_sz)
 {
   auto start = std::chrono::high_resolution_clock::now();
-  Network net ("./data_banknote_authentication.txt", batch_sz, 0.0155, 0.03, 0, 0.9);
+  Network net ("./data_banknote_authentication.txt", batch_sz, 0.1, 0.03, 0, 0.9);
   net.add_layer(4, "linear");
   net.add_layer(6, "lecun_tanh");
   net.add_layer(2, "linear");
@@ -41,7 +41,6 @@ double bench(int batch_sz)
 
 int main()
 {
-  //  sleep(30);
   std::cout << bench(16) << "\n";
   //  bench(50);
   //  bench(50);
diff --git a/src/bpnn.cpp b/src/bpnn.cpp
@@ -146,21 +146,7 @@ void Network::set_activation(int index, std::function<float(float)> custom, std:
 
 void Network::feedforward()
 {
-  for (int j = 0; j < layers[0].contents->rows(); j++) {
-    if (strcmp(layers[0].activation_str, "linear") == 0) break;
-    for (int k = 0; k < layers[0].contents->cols(); k++) {
-      (*layers[0].dZ)(j,k) = layers[0].activation_deriv((*layers[0].contents)(j,k));
-      (*layers[0].contents)(j,k) = layers[0].activation((*layers[0].contents)(j,k));
-    }
-  }
   for (int i = 0; i < length-1; i++) {
-    //if (batch_size > 64 && batch_size % 4 == 0) {
-    //  *layers[i+1].contents = strassen_mul((*layers[i].contents),(*layers[i].weights));
-    //}
-    *layers[i+1].contents = (*layers[i].contents) * (*layers[i].weights);
-    *layers[i+1].contents += *layers[i+1].bias;
-  }
-  for (int i = 1; i < length; i++) {
     for (int j = 0; j < layers[i].contents->rows(); j++) {
       if (strcmp(layers[i].activation_str, "linear") == 0) break;
       for (int k = 0; k < layers[i].contents->cols(); k++) {
@@ -168,6 +154,15 @@ void Network::feedforward()
         (*layers[i].contents)(j,k) = layers[i].activation((*layers[i].contents)(j,k));
       }
     }
+    *layers[i+1].contents = (*layers[i].contents) * (*layers[i].weights);
+    *layers[i+1].contents += *layers[i+1].bias;
+  }
+  for (int j = 0; j < layers[length-1].contents->rows(); j++) {
+    if (strcmp(layers[length-1].activation_str, "linear") == 0) break;
+    for (int k = 0; k < layers[length-1].contents->cols(); k++) {
+      (*layers[length-1].dZ)(j,k) = layers[length-1].activation_deriv((*layers[length-1].contents)(j,k));
+      (*layers[length-1].contents)(j,k) = layers[length-1].activation((*layers[length-1].contents)(j,k));
+    }
   }
   //  std::cout << "\nSOFTMAX INPUT\n" << *layers[length-1].contents << "\n\n";
   for (int i = 0; i < layers[length-1].contents->rows(); i++) {