linear2d_layer: make it 2d

Author: OneAdder

src/nf/nf_linear2d_layer.f90

@@ -13,8 +13,8 @@ module nf_linear2d_layer
 
     real, allocatable :: weights(:, :)
     real, allocatable :: biases(:)
-    real, allocatable :: output(:, :, :)
-    real, allocatable :: gradient(:, :, :) ! input gradient
+    real, allocatable :: output(:, :)
+    real, allocatable :: gradient(:, :) ! input gradient
     real, allocatable :: dw(:, :) ! weight gradients
     real, allocatable :: db(:) ! bias gradients
 
@@ -32,23 +32,23 @@ module nf_linear2d_layer
 
   interface linear2d_layer
     module function linear2d_layer_cons(&
-        sequence_length, in_features, out_features, batch_size&
+        sequence_length, in_features, out_features&
     ) result(res)
-      integer, intent(in) :: batch_size, sequence_length, in_features, out_features
+      integer, intent(in) :: sequence_length, in_features, out_features
       type(linear2d_layer) :: res
     end function linear2d_layer_cons
   end interface linear2d_layer
 
   interface
     pure module subroutine forward(self, input)
       class(linear2d_layer), intent(in out) :: self
-      real, intent(in) :: input(:, :, :)
+      real, intent(in) :: input(:, :)
     end subroutine forward
 
     pure module subroutine backward(self, input, gradient)
       class(linear2d_layer), intent(in out) :: self
-      real, intent(in) :: input(:, :, :)
-      real, intent(in) :: gradient(:, :, :)
+      real, intent(in) :: input(:, :)
+      real, intent(in) :: gradient(:, :)
     end subroutine backward
 
     module subroutine init(self, input_shape)

test/test_linear2d_layer.f90

@@ -4,14 +4,13 @@ program test_linear2d_layer
   implicit none
 
   logical :: ok = .true.
-  real :: sample_input(3, 4, 2) = reshape(&
-      [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.1,&
-       0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2],&
-      [3, 4, 2]) ! first batch are 0.1, second 0.2
-  real :: sample_gradient(3, 1, 2) = reshape([2., 2., 2., 3., 3., 3.], [3, 1, 2])
+  real :: sample_input(3, 4) = reshape(&
+      [0.1, 0.1, 0.1, 0.1, 0.1, 0.1, 0.2, 0.2, 0.2, 0.2, 0.2, 0.2],&
+      [3, 4]) ! first batch are 0.1, second 0.2
+  real :: sample_gradient(3, 1) = reshape([2., 2., 3.], [3, 1])
   type(linear2d_layer) :: linear
 
-  linear = linear2d_layer(sequence_length=3, in_features=4, out_features=1, batch_size=2)
+  linear = linear2d_layer(sequence_length=3, in_features=4, out_features=1)
   call linear % init([4])
 
   call test_linear2d_layer_forward(linear, ok, sample_input)
@@ -22,11 +21,11 @@ program test_linear2d_layer
   subroutine test_linear2d_layer_forward(linear, ok, input)
     type(linear2d_layer), intent(in out) :: linear
     logical, intent(in out) :: ok
-    real, intent(in) :: input(3, 4, 2)
-    real :: output_shape(3)
-    real :: output_flat(6)
-    real :: expected_shape(3) = [3, 1, 2]
-    real :: expected_output_flat(6) = [0.15, 0.15, 0.15, 0.19, 0.19, 0.19]
+    real, intent(in) :: input(3, 4)
+    real :: output_shape(2)
+    real :: output_flat(3)
+    real :: expected_shape(2) = [3, 1]
+    real :: expected_output_flat(3) = [0.17, 0.17, 0.17]
 
     call linear % forward(input)
 
@@ -45,28 +44,23 @@ end subroutine test_linear2d_layer_forward
   subroutine test_linear2d_layer_backward(linear, ok, input, gradient)
     type(linear2d_layer), intent(in out) :: linear
     logical, intent(in out) :: ok
-    real, intent(in) :: input(3, 4, 2)
-    real, intent(in) :: gradient(3, 1, 2)
-    real :: gradient_shape(3)
+    real, intent(in) :: input(3, 4)
+    real, intent(in) :: gradient(3, 1)
+    real :: gradient_shape(2)
     real :: dw_shape(2)
     real :: db_shape(1)
-    real :: gradient_flat(24)
+    real :: gradient_flat(12)
     real :: dw_flat(4)
-    real :: expected_gradient_shape(3) = [3, 4, 2]
+    real :: expected_gradient_shape(2) = [3, 4]
     real :: expected_dw_shape(2) = [4, 1]
     real :: expected_db_shape(1) = [1]
-    real :: expected_gradient_flat(24) = [&
-        0.200000003, 0.200000003, 0.200000003, 0.200000003,&
-        0.200000003, 0.200000003, 0.200000003, 0.200000003,&
-        0.200000003, 0.200000003, 0.200000003, 0.200000003,&
-        0.300000012, 0.300000012, 0.300000012, 0.300000012,&
-        0.300000012, 0.300000012, 0.300000012, 0.300000012,&
-        0.300000012, 0.300000012, 0.300000012, 0.300000012&
+    real :: expected_gradient_flat(12) = [&
+        0.2, 0.2, 0.3, 0.2,&
+        0.2, 0.3, 0.2, 0.2,&
+        0.3, 0.2, 0.2, 0.3&
     ]
-    real :: expected_dw_flat(4)
-    real :: expected_db(1) = [15.0]
-
-    expected_dw_flat = 2.40000010
+    real :: expected_dw_flat(4) = [0.7, 0.7, 1.4, 1.4]
+    real :: expected_db(1) = [7]
 
     call linear % backward(input, gradient)
 
@@ -104,8 +98,8 @@ end subroutine test_linear2d_layer_backward
 
   subroutine test_linear2d_layer_gradient_updates(ok)
     logical, intent(in out) :: ok
-    real :: input(3, 4, 1) = reshape([0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.11, 0.12], [3, 4, 1])
-    real :: gradient(3, 2, 1) = reshape([0.0, 10., 0.2, 3., 0.4, 1.], [3, 2, 1])
+    real :: input(3, 4) = reshape([0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 0.11, 0.12], [3, 4])
+    real :: gradient(3, 2) = reshape([0.0, 10., 0.2, 3., 0.4, 1.], [3, 2])
     type(linear2d_layer) :: linear
 
     integer :: num_parameters