Distinguish between testing and training when optimizing the network model. This is useful for models that act differently when in testing or training mode.

src/mlpack/methods/ann/cnn.hpp

@@ -182,7 +182,7 @@ class CNN
    */
   double Evaluate(const arma::mat& parameters,
                   const size_t i,
-                  const bool deterministic = false);
+                  const bool deterministic = true);
 
   /**
    * Evaluate the gradient of the convolutional neural network with the given

src/mlpack/methods/ann/cnn_impl.hpp

@@ -245,9 +245,11 @@ template<typename LayerTypes,
 void CNN<
 LayerTypes, OutputLayerType, InitializationRuleType, PerformanceFunction
 >::Gradient(const arma::mat& /* unused */,
-            const size_t /* unused */,
+            const size_t i,
             arma::mat& gradient)
 {
+  Evaluate(parameter, i, false);
+
   NetworkGradients(gradient, network);
 
   Backward<>(error, network);

src/mlpack/methods/ann/ffn.hpp

@@ -183,7 +183,7 @@ class FFN
    */
   double Evaluate(const arma::mat& parameters,
                   const size_t i,
-                  const bool deterministic = false);
+                  const bool deterministic = true);
 
   /**
    * Evaluate the gradient of the feedforward network with the given parameters,

src/mlpack/methods/ann/ffn_impl.hpp

@@ -249,9 +249,11 @@ template<typename LayerTypes,
 void FFN<
 LayerTypes, OutputLayerType, InitializationRuleType, PerformanceFunction
 >::Gradient(const arma::mat& /* unused */,
-            const size_t /* unused */,
+            const size_t i,
             arma::mat& gradient)
 {
+  Evaluate(parameter, i, false);
+
   NetworkGradients(gradient, network);
 
   Backward<>(error, network);

src/mlpack/methods/ann/rnn.hpp

@@ -185,7 +185,7 @@ class RNN
    */
   double Evaluate(const arma::mat& parameters,
                   const size_t i,
-                  const bool deterministic = false);
+                  const bool deterministic = true);
 
   /**
    * Evaluate the gradient of the recurrent neural network with the given

src/mlpack/methods/ann/rnn_impl.hpp

@@ -281,6 +281,8 @@ LayerTypes, OutputLayerType, InitializationRuleType, PerformanceFunction
             const size_t i,
             arma::mat& gradient)
 {
+  Evaluate(parameter, i, false);
+
   gradient.zeros();
   arma::mat currentGradient = arma::mat(gradient.n_rows, gradient.n_cols);
   NetworkGradients(currentGradient, network);