TrainableLayer is now a concept instead of inehritable object: nim-la…

…ng/Nim#7713

src/nn_dsl/dsl_types.nim

@@ -55,13 +55,18 @@ type
 
   #####################################################
   # Todo: move that in NN part
-  TrainableLayer*[TT] = object of RootObj
+  # We have to use concepts here: non-ref object inheritance
+  # doesn't work properly: https://github.com/nim-lang/Nim/issues/7713
+  TrainableLayer*[TT] = concept layer
+    layer.weight is Variable[TT]
+    layer.bias   is Variable[TT]
+
+  Conv2DLayer*[TT] = object
+    weight*: Variable[TT]
+    bias*: Variable[TT]
+  LinearLayer*[TT] = object
     weight*: Variable[TT]
     bias*: Variable[TT]
-
-  Conv2DLayer*[T] = object of TrainableLayer[T]
-  LinearLayer*[T] = object of TrainableLayer[T]
-
 
 proc hash*(x: NimNode): Hash =
   assert x.kind == nnkIdent

src/nn_dsl/nn_dsl.nim

@@ -2,7 +2,10 @@
 # Distributed under the Apache v2 License (license terms are at http://www.apache.org/licenses/LICENSE-2.0).
 # This file may not be copied, modified, or distributed except according to those terms.
 
-import dsl_core, dsl_types
+import
+  macros,
+  dsl_core, dsl_types,
+  ../nn/nn
 
 export
   network,
@@ -20,3 +23,13 @@ proc flatten*(s: openarray[int]): int {.inline.}=
   result = 1
   for val in s:
     result *= val
+
+func optimizer*[M; U: SomeReal](model: M, optimizer: typedesc[Optimizer[U]], learning_rate: U): Optimizer[U] =
+  result.params = @[]
+  result.lr = learning_rate
+
+  for layer in fields(model):
+    when layer is TrainableLayer:
+      result.params.add layer.weight
+      result.params.add layer.bias
+