Issue error when serializing ANN models without MLPACK_ENABLE_ANN_SERIALIZATION

HISTORY.md

@@ -5,7 +5,8 @@
 
   * Adapt PReLU layer for new neural network API (#3420).
 
-  * Add CF decomposition methods: `QUIC_SVDPolicy` and `BlockKrylovSVDPolicy` (#3413, #3404).
+  * Add CF decomposition methods: `QUIC_SVDPolicy` and `BlockKrylovSVDPolicy`
+    (#3413, #3404).
 
   * Update outdated code in tutorials (#3398, #3401).
 
@@ -17,6 +18,9 @@
 
   * Avoid deprecation warnings in Armadillo 11.4.4+ (#3405).
 
+  * Issue runtime error when serialization of neural networks is attempted but
+    `MLPACK_ENABLE_ANN_SERIALIZATION` is not defined (#3451).
+
 ### mlpack 4.0.1
 ###### 2022-12-23
   * Fix mapping of categorical data for Julia bindings (#3305).

README.md

@@ -180,6 +180,8 @@ g++ -O3 -std=c++14 -o my_program my_program.cpp -larmadillo -fopenmp
 
 Note that if you want to serialize (save or load) neural networks, you should
 add `#define MLPACK_ENABLE_ANN_SERIALIZATION` before including `<mlpack.hpp>`.
+If you don't define `MLPACK_ENABLE_ANN_SERIALIZATION` and your code serializes a
+neural network, a compilation error will occur.
 
 See the [C++ quickstart](doc/quickstart/cpp.md) and the
 [examples](https://github.com/mlpack/examples) repository for some examples of
@@ -198,7 +200,8 @@ reduce compilation time:
  * Only use the `MLPACK_ENABLE_ANN_SERIALIZATION` definition if you are
    serializing neural networks in your code.  When this define is enabled,
    compilation time will increase significantly, as the compiler must generate
-   code for every possible type of layer.
+   code for every possible type of layer.  (The large amount of extra
+   compilation overhead is why this is not enabled by default.)
 
  * If you are using mlpack in multiple .cpp files, consider using [`extern
    templates`](https://isocpp.org/wiki/faq/cpp11-language-templates) so that the

src/mlpack/methods/ann/ffn_impl.hpp

@@ -374,34 +374,46 @@ void FFN<
     MatType
 >::serialize(Archive& ar, const uint32_t /* version */)
 {
-  // Serialize the output layer and initialization rule.
-  ar(CEREAL_NVP(outputLayer));
-  ar(CEREAL_NVP(initializeRule));
-
-  // Serialize the network itself.
-  ar(CEREAL_NVP(network));
-  ar(CEREAL_NVP(parameters));
-
-  // Serialize the expected input size.
-  ar(CEREAL_NVP(inputDimensions));
-
-  // If we are loading, we need to initialize the weights.
-  if (cereal::is_loading<Archive>())
-  {
-    // We can clear these members, since it's not possible to serialize in the
-    // middle of training and resume.
-    predictors.clear();
-    responses.clear();
-
-    networkOutput.clear();
-    networkDelta.clear();
-
-    layerMemoryIsSet = false;
-    inputDimensionsAreSet = false;
-
-    // The weights in `parameters` will be correctly set for each layer in the
-    // first call to Forward().
-  }
+  #ifndef MLPACK_ENABLE_ANN_SERIALIZATION
+    // Note: if you define MLPACK_IGNORE_ANN_SERIALIZATION_WARNING, you had
+    // better ensure that every layer you are serializing has had
+    // CEREAL_REGISTER_TYPE() called somewhere.  See layer/serialization.hpp for
+    // more information.
+    #ifndef MLPACK_ANN_IGNORE_SERIALIZATION_WARNING
+      throw std::runtime_error("Cannot serialize a neural network unless "
+          "MLPACK_ENABLE_ANN_SERIALIZATION is defined!  See the \"Additional "
+          "build options\" section of the README for more information.");
+    #endif
+  #else
+    // Serialize the output layer and initialization rule.
+    ar(CEREAL_NVP(outputLayer));
+    ar(CEREAL_NVP(initializeRule));
+
+    // Serialize the network itself.
+    ar(CEREAL_NVP(network));
+    ar(CEREAL_NVP(parameters));
+
+    // Serialize the expected input size.
+    ar(CEREAL_NVP(inputDimensions));
+
+    // If we are loading, we need to initialize the weights.
+    if (cereal::is_loading<Archive>())
+    {
+      // We can clear these members, since it's not possible to serialize in the
+      // middle of training and resume.
+      predictors.clear();
+      responses.clear();
+
+      networkOutput.clear();
+      networkDelta.clear();
+
+      layerMemoryIsSet = false;
+      inputDimensionsAreSet = false;
+
+      // The weights in `parameters` will be correctly set for each layer in the
+      // first call to Forward().
+    }
+  #endif
 }
 
 template<typename OutputLayerType,

src/mlpack/methods/ann/rnn_impl.hpp

@@ -290,17 +290,29 @@ void RNN<
 >::serialize(
     Archive& ar, const uint32_t /* version */)
 {
-  ar(CEREAL_NVP(bpttSteps));
-  ar(CEREAL_NVP(single));
-  ar(CEREAL_NVP(network));
-
-  if (Archive::is_loading::value)
-  {
-    // We can clear these members, since it's not possible to serialize in the
-    // middle of training and resume.
-    predictors.clear();
-    responses.clear();
-  }
+  #ifndef MLPACK_ENABLE_ANN_SERIALIZATION
+    // Note: if you define MLPACK_IGNORE_ANN_SERIALIZATION_WARNING, you had
+    // better ensure that every layer you are serializing has had
+    // CEREAL_REGISTER_TYPE() called somewhere.  See layer/serialization.hpp for
+    // more information.
+    #ifndef MLPACK_IGNORE_ANN_SERIALIZATION_WARNING
+      throw std::runtime_error("Cannot serialize a neural network unless "
+          "MLPACK_ENABLE_ANN_SERIALIZATION is defined!  See the \"Additional "
+          "build options\" section of the README for more information.");
+    #endif
+  #else
+    ar(CEREAL_NVP(bpttSteps));
+    ar(CEREAL_NVP(single));
+    ar(CEREAL_NVP(network));
+
+    if (Archive::is_loading::value)
+    {
+      // We can clear these members, since it's not possible to serialize in the
+      // middle of training and resume.
+      predictors.clear();
+      responses.clear();
+    }
+  #endif
 }
 
 template<

src/mlpack/tests/CMakeLists.txt

@@ -221,7 +221,14 @@ else()
   target_compile_definitions(mlpack_test PUBLIC -DMLPACK_SUPPRESS_FATAL)
 endif()
 
-set_target_properties(mlpack_test PROPERTIES COTIRE_CXX_PREFIX_HEADER_INIT "../core.hpp")
+# This has to be added here so that cotire picks it up (even though it is in
+# individual tests).
+target_compile_definitions(mlpack_test PUBLIC -DMLPACK_ENABLE_ANN_SERIALIZATION)
+set_target_properties(mlpack_test PROPERTIES COTIRE_CXX_PREFIX_HEADER_INIT
+    "../core.hpp")
+# TODO: use the source below, but this requires the DET test to be refactored
+# and cleaned up.
+#    "../../mlpack.hpp")
 cotire(mlpack_test)
 
 # Copy test data into right place.

src/mlpack/tests/ann/feedforward_network_test.cpp

@@ -10,7 +10,9 @@
  * 3-clause BSD license along with mlpack.  If not, see
  * http://www.opensource.org/licenses/BSD-3-Clause for more information.
  */
-#define MLPACK_ENABLE_ANN_SERIALIZATION
+#ifndef MLPACK_ENABLE_ANN_SERIALIZATION
+  #define MLPACK_ENABLE_ANN_SERIALIZATION
+#endif
 #include <mlpack/core.hpp>
 #include <mlpack/methods/ann/ann.hpp>
 #include <mlpack/methods/kmeans/kmeans.hpp>

src/mlpack/tests/ann/layer/batch_norm.cpp

@@ -10,7 +10,9 @@
  * 3-clause BSD license along with mlpack.  If not, see
  * http://www.opensource.org/licenses/BSD-3-Clause for more information.
  */
-#define MLPACK_ENABLE_ANN_SERIALIZATION
+#ifndef MLPACK_ENABLE_ANN_SERIALIZATION
+  #define MLPACK_ENABLE_ANN_SERIALIZATION
+#endif
 #include <mlpack/core.hpp>
 #include <mlpack/methods/ann.hpp>
 

src/mlpack/tests/main.cpp

@@ -8,6 +8,9 @@
  * 3-clause BSD license along with mlpack.  If not, see
  * http://www.opensource.org/licenses/BSD-3-Clause for more information.
  */
+#ifndef MLPACK_ENABLE_ANN_SERIALIZATION
+  #define MLPACK_ENABLE_ANN_SERIALIZATION
+#endif
 #include <mlpack.hpp>
 
 // #define CATCH_CONFIG_MAIN  // catch.hpp will define main()