Merge pull request #2650 from UtR491/issue-2590

Added Weight Size to concat and fast_lstm layers and

src/mlpack/methods/ann/layer/add_impl.hpp

@@ -23,7 +23,7 @@ template<typename InputDataType, typename OutputDataType>
 Add<InputDataType, OutputDataType>::Add(const size_t outSize) :
     outSize(outSize)
 {
-  weights.set_size(outSize, 1);
+  weights.set_size(WeightSize(), 1);
 }
 
 template<typename InputDataType, typename OutputDataType>

src/mlpack/methods/ann/layer/atrous_convolution_impl.hpp

@@ -122,8 +122,7 @@ AtrousConvolution<
     dilationWidth(dilationWidth),
     dilationHeight(dilationHeight)
 {
-  weights.set_size((outSize * inSize * kernelWidth * kernelHeight) + outSize,
-      1);
+  weights.set_size(WeightSize(), 1);
 
   // Transform paddingType to lowercase.
   std::string paddingTypeLow = paddingType;

src/mlpack/methods/ann/layer/concat.hpp

@@ -165,9 +165,9 @@ class Concat
   }
 
   //! Return the initial point for the optimization.
-  const arma::mat& Parameters() const { return parameters; }
+  const arma::mat& Parameters() const { return weights; }
   //! Modify the initial point for the optimization.
-  arma::mat& Parameters() { return parameters; }
+  arma::mat& Parameters() { return weights; }
 
   //! Get the value of run parameter.
   bool Run() const { return run; }
@@ -196,6 +196,9 @@ class Concat
   //! Get the axis of concatenation.
   size_t const& ConcatAxis() const { return axis; }
 
+  //! Get the size of the weight matrix.
+  size_t WeightSize() const { return 0; }
+
   /**
    * Serialize the layer
    */
@@ -225,8 +228,8 @@ class Concat
   //! Locally-stored network modules.
   std::vector<LayerTypes<CustomLayers...> > network;
 
-  //! Locally-stored model parameters.
-  arma::mat parameters;
+  //! Locally-stored model weights.
+  OutputDataType weights;
 
   //! Locally-stored delta visitor.
   DeltaVisitor deltaVisitor;

src/mlpack/methods/ann/layer/concat_impl.hpp

@@ -33,7 +33,7 @@ Concat<InputDataType, OutputDataType, CustomLayers...>::Concat(
     run(run),
     channels(1)
 {
-  parameters.set_size(0, 0);
+  weights.set_size(0, 0);
 }
 
 template<typename InputDataType, typename OutputDataType,
@@ -49,7 +49,7 @@ Concat<InputDataType, OutputDataType, CustomLayers...>::Concat(
     model(model),
     run(run)
 {
-  parameters.set_size(0, 0);
+  weights.set_size(0, 0);
 
   // Parameters to help calculate the number of channels.
   size_t oldColSize = 1, newColSize = 1;

src/mlpack/methods/ann/layer/dropconnect.hpp

@@ -115,9 +115,9 @@ class DropConnect
   std::vector<LayerTypes<> >& Model() { return network; }
 
   //! Get the parameters.
-  OutputDataType const& Parameters() const { return parameters; }
+  OutputDataType const& Parameters() const { return weights; }
   //! Modify the parameters.
-  OutputDataType& Parameters() { return parameters; }
+  OutputDataType& Parameters() { return weights; }
 
   //! Get the output parameter.
   OutputDataType const& OutputParameter() const { return outputParameter; }
@@ -150,6 +150,9 @@ class DropConnect
     scale = 1.0 / (1.0 - ratio);
   }
 
+  //! Return the size of the weight matrix.
+  size_t WeightSize() const { return 0; }
+
   /**
    * Serialize the layer.
    */
@@ -164,7 +167,7 @@ class DropConnect
   double scale;
 
   //! Locally-stored weight object.
-  OutputDataType parameters;
+  OutputDataType weights;
 
   //! Locally-stored delta object.
   OutputDataType delta;

src/mlpack/methods/ann/layer/fast_lstm.hpp

@@ -164,6 +164,12 @@ class FastLSTM
   //! Get the number of output units.
   size_t OutSize() const { return outSize; }
 
+  //! Get the size of the weight matrix.
+  size_t WeightSize() const 
+  {
+    return 4 * outSize * inSize + 4 * outSize + 4 * outSize * outSize;
+  }
+
   /**
    * Serialize the layer
    */

src/mlpack/methods/ann/layer/fast_lstm_impl.hpp

@@ -42,8 +42,7 @@ FastLSTM<InputDataType, OutputDataType>::FastLSTM(
 {
   // Weights for: input to gate layer (4 * outsize * inSize + 4 * outsize)
   // and output to gate (4 * outSize).
-  weights.set_size(
-      4 * outSize * inSize + 4 * outSize + 4 * outSize * outSize, 1);
+  weights.set_size(WeightSize(), 1);
 }
 
 template<typename InputDataType, typename OutputDataType>

src/mlpack/methods/ann/layer/linear_impl.hpp

@@ -38,7 +38,7 @@ Linear<InputDataType, OutputDataType, RegularizerType>::Linear(
     outSize(outSize),
     regularizer(regularizer)
 {
-  weights.set_size(outSize * inSize + outSize, 1);
+  weights.set_size(WeightSize(), 1);
 }
 
 template<typename InputDataType, typename OutputDataType,

src/mlpack/tests/ann_visitor_test.cpp

@@ -86,15 +86,78 @@ TEST_CASE("WeightSetVisitorTest", "[ANNVisitorTest]")
 }
 
 /**
- * Test that WeightSizeVisitor works properly.
+ * Test that WeightSizeVisitor works properly for linear layer.
  */
-TEST_CASE("WeightSizeVisitorTest", "[ANNVisitorTest]")
+TEST_CASE("WeightSizeVisitorTestForLinearLayer", "[ANNVisitorTest]")
 {
-  size_t randomSize = arma::randi(arma::distr_param(1, 100));
+  size_t randomInSize = arma::randi(arma::distr_param(1, 100));
+  size_t randomOutSize = arma::randi(arma::distr_param(1, 100));
 
-  LayerTypes<> linear = new Linear<>(randomSize, randomSize);
+  LayerTypes<> linearLayer = new Linear<>(randomInSize, randomOutSize);
 
-  CheckCorrectnessOfWeightSize(linear);
+  size_t weightSize = boost::apply_visitor(WeightSizeVisitor(), linearLayer);
+
+  CheckCorrectnessOfWeightSize(linearLayer);
+}
+
+/**
+ * Test that WeightSizeVisitor works properly for concat layer.
+ */
+TEST_CASE("WeightSizeVisitorTestForConcatLayer", "[ANNVisitorTest]")
+{
+  LayerTypes<> concatLayer = new Concat<>();
+  
+  size_t weightSize = boost::apply_visitor(WeightSizeVisitor(), concatLayer);
+
+  CheckCorrectnessOfWeightSize(concatLayer);
+}
+
+/**
+ * Test that WeightSizeVisitor works properly for fast lstm layer.
+ */
+TEST_CASE("WeightSizeVisitorTestForFastLSTMLayer", "[ANNVisitorTest]")
+{
+  size_t randomInSize = arma::randi(arma::distr_param(1, 100));
+  size_t randomOutSize = arma::randi(arma::distr_param(1, 100));
+
+  LayerTypes<> fastLSTMLayer = new FastLSTM<>(randomInSize, randomOutSize);
+
+  size_t weightSize = boost::apply_visitor(WeightSizeVisitor(), fastLSTMLayer);
+
+  CheckCorrectnessOfWeightSize(fastLSTMLayer);
+}
+
+/**
+ * Test that WeightSizeVisitor works properly for Add layer.
+ */
+TEST_CASE("WeightSizeVisitorTestForAddLayer", "[ANNVisitorTest]")
+{
+  size_t randomOutSize = arma::randi(arma::distr_param(1, 100));
+
+  LayerTypes<> addLayer = new Add<>(randomOutSize);
+
+  size_t weightSize = boost::apply_visitor(WeightSizeVisitor(), addLayer);
+
+  CheckCorrectnessOfWeightSize(addLayer);
+}
+
+/**
+ * Test that WeightSizeVisitor works properly for Atrous Convolution Layer.
+ */
+TEST_CASE("WeightSizeVisitorTestForAtrousConvolutionLayer", "[ANNVisitorTest]")
+{
+  size_t randomInSize = arma::randi(arma::distr_param(1, 100));
+  size_t randomOutSize = arma::randi(arma::distr_param(1, 100));
+  size_t randomKernelWidth = arma::randi(arma::distr_param(1, 100));
+  size_t randomKernelHeight = arma::randi(arma::distr_param(1, 100));
+
+  LayerTypes<> atrousConvLayer = new AtrousConvolution<>(randomInSize, randomOutSize,
+      randomKernelWidth, randomKernelHeight);
+
+  size_t weightSize = boost::apply_visitor(WeightSizeVisitor(),
+      atrousConvLayer);
+ 
+  CheckCorrectnessOfWeightSize(atrousConvLayer);
 }