Locally connected layer

CMakeLists.txt

@@ -36,6 +36,8 @@ add_library(neural-fortran
   src/nf/nf_layer_constructors_submodule.f90
   src/nf/nf_layer.f90
   src/nf/nf_layer_submodule.f90
+  src/nf/nf_locally_connected_1d.f90
+  src/nf/nf_locally_connected_1d_submodule.f90
   src/nf/nf_loss.f90
   src/nf/nf_loss_submodule.f90
   src/nf/nf_maxpool2d_layer.f90
@@ -47,6 +49,8 @@ add_library(neural-fortran
   src/nf/nf_parallel.f90
   src/nf/nf_parallel_submodule.f90
   src/nf/nf_random.f90
+  src/nf/nf_reshape_generalized.f90
+  src/nf/nf_reshape_generalized_submodule.f90
   src/nf/nf_reshape_layer.f90
   src/nf/nf_reshape_layer_submodule.f90
   src/nf/io/nf_io_binary.f90

example/CMakeLists.txt

@@ -1,5 +1,6 @@
 foreach(execid
   cnn_mnist
+  cnn_mnist_1d
   dense_mnist
   get_set_network_params
   network_parameters

example/cnn_mnist.f90

@@ -12,7 +12,7 @@ program cnn_mnist
   real, allocatable :: validation_images(:,:), validation_labels(:)
   real, allocatable :: testing_images(:,:), testing_labels(:)
   integer :: n
-  integer, parameter :: num_epochs = 10
+  integer, parameter :: num_epochs = 20
 
   call load_mnist(training_images, training_labels, &
                   validation_images, validation_labels, &
@@ -35,9 +35,9 @@ program cnn_mnist
     call net % train( &
       training_images, &
       label_digits(training_labels), &
-      batch_size=128, &
+      batch_size=16, &
       epochs=1, &
-      optimizer=sgd(learning_rate=3.) &
+      optimizer=sgd(learning_rate=0.003) &
     )
 
     print '(a,i2,a,f5.2,a)', 'Epoch ', n, ' done, Accuracy: ', accuracy( &

example/cnn_mnist_1d.f90

@@ -0,0 +1,64 @@
+program cnn_mnist
+
+    use nf, only: network, sgd, &
+      input, conv2d, maxpool2d, flatten, dense, reshape, reshape_generalized, locally_connected_1d, &
+      load_mnist, label_digits, softmax, relu
+
+    implicit none
+
+    type(network) :: net
+
+    real, allocatable :: training_images(:,:), training_labels(:)
+    real, allocatable :: validation_images(:,:), validation_labels(:)
+    real, allocatable :: testing_images(:,:), testing_labels(:)
+    integer :: n
+    integer, parameter :: num_epochs = 10
+
+    call load_mnist(training_images, training_labels, &
+                    validation_images, validation_labels, &
+                    testing_images, testing_labels)
+
+    net = network([ &
+      input(784), &
+      reshape_generalized([28, 28]), &
+      locally_connected_1d(filters=8, kernel_size=3, activation=relu()), &
+      dense(10, activation=softmax()) &
+    ])
+
+    call net % print_info()
+
+    epochs: do n = 1, num_epochs
+
+      call net % train( &
+        training_images, &
+        label_digits(training_labels), &
+        batch_size=16, &
+        epochs=1, &
+        optimizer=sgd(learning_rate=1) &
+      )
+
+      print '(a,i2,a,f5.2,a)', 'Epoch ', n, ' done, Accuracy: ', accuracy( &
+        net, validation_images, label_digits(validation_labels)) * 100, ' %'
+
+    end do epochs
+
+    print '(a,f5.2,a)', 'Testing accuracy: ', &
+      accuracy(net, testing_images, label_digits(testing_labels)) * 100, '%'
+
+  contains
+
+    real function accuracy(net, x, y)
+      type(network), intent(in out) :: net
+      real, intent(in) :: x(:,:), y(:,:)
+      integer :: i, good
+      good = 0
+      do i = 1, size(x, dim=2)
+        if (all(maxloc(net % predict(x(:,i))) == maxloc(y(:,i)))) then
+          good = good + 1
+        end if
+      end do
+      accuracy = real(good) / size(x, dim=2)
+    end function accuracy
+
+  end program cnn_mnist
+

src/nf.f90

@@ -3,7 +3,7 @@ module nf
   use nf_datasets_mnist, only: label_digits, load_mnist
   use nf_layer, only: layer
   use nf_layer_constructors, only: &
-    conv2d, dense, flatten, input, maxpool2d, reshape
+    conv2d, dense, flatten, input, maxpool2d, reshape, reshape_generalized, locally_connected_1d
   use nf_loss, only: mse, quadratic
   use nf_metrics, only: corr, maxabs
   use nf_network, only: network