Constructor for RNN

Author: castelao

src/nf/nf_layer_constructors.f90

@@ -166,6 +166,29 @@ pure module function reshape(output_shape) result(res)
         !! Resulting layer instance
     end function reshape
 
+    pure module function rnn(layer_size, activation) result(res)
+      !! Recurrent (fully-connected) layer constructor.
+      !!
+      !! This layer is a building block for recurrent, fully-connected
+      !! networks, or for an output layer of a convolutional network.
+      !! A recurrent layer must not be the first layer in the network.
+      !!
+      !! Example:
+      !!
+      !! ```
+      !! use nf, only :: rnn, layer, relu
+      !! type(layer) :: rnn_layer
+      !! rnn_layer = rnn(10)
+      !! rnn_layer = rnn(10, activation=relu())
+      !! ```
+      integer, intent(in) :: layer_size
+        !! The number of neurons in a dense layer
+      class(activation_function), intent(in), optional :: activation
+        !! Activation function instance (default tanh)
+      type(layer) :: res
+        !! Resulting layer instance
+    end function rnn
+
   end interface
 
 end module nf_layer_constructors

src/nf/nf_layer_constructors_submodule.f90

@@ -9,7 +9,7 @@
   use nf_maxpool2d_layer, only: maxpool2d_layer
   use nf_reshape_layer, only: reshape3d_layer
   use nf_rnn_layer, only: rnn_layer
-  use nf_activation, only: activation_function, relu, sigmoid
+  use nf_activation, only: activation_function, relu, sigmoid, tanhf
 
   implicit none
 
@@ -135,4 +135,27 @@ pure module function reshape(output_shape) result(res)
 
   end function reshape
 
+  pure module function rnn(layer_size, activation) result(res)
+    integer, intent(in) :: layer_size
+    class(activation_function), intent(in), optional :: activation
+    type(layer) :: res
+
+    class(activation_function), allocatable :: activation_tmp
+
+    res % name = 'rnn'
+    res % layer_shape = [layer_size]
+
+    if (present(activation)) then
+      allocate(activation_tmp, source=activation)
+    else
+      allocate(activation_tmp, source=tanhf())
+    end if
+
+    res % activation = activation_tmp % get_name()
+
+    allocate(res % p, source=rnn_layer(layer_size, activation_tmp))
+
+  end function rnn
+
+
 end submodule nf_layer_constructors_submodule