Merge branch 'master' of https://github.com/mrdaybird/mlpack into fix_logsoftmax

HISTORY.md

@@ -1,13 +1,17 @@
 ### mlpack ?.?.?
 ###### ????-??-??
 
+### mlpack 4.1.0
+###### 2023-04-26
+
   * Adapt HardTanH layer (#3454).
 
   * Adapt Softmin layer for new neural network API (#3437).
 
   * 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   * Update outdated code in tutorials (#3398, #3401).
 
@@ -19,6 +23,9 @@
 
   * 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

@@ -21,7 +21,7 @@ src="https://cdn.rawgit.com/mlpack/mlpack.org/e7d36ed8/mlpack-black.svg" style="
 <p align="center">
   <em>
     Download:
-    <a href="https://www.mlpack.org/files/mlpack-4.0.1.tar.gz">current stable version (4.0.1)</a>
+    <a href="https://www.mlpack.org/files/mlpack-4.1.0.tar.gz">current stable version (4.1.0)</a>
   </em>
 </p>
 
@@ -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
@@ -252,7 +255,7 @@ build in parallel; e.g., `make -j4` will use 4 cores to build.
 
 mlpack's Python bindings are available on
 [PyPI](https://pypi.org/project/mlpack) and
-[conda-forge](https://conda-forge.org/packages/mlpack), and can be installed
+[conda-forge](https://anaconda.org/conda-forge/mlpack), and can be installed
 with either `pip install mlpack` or `conda install -c conda-forge mlpack`.
 These sources are recommended, as building the Python bindings by hand can be
 complex.

doc/examples/sample-ml-app/sample-ml-app/sample-ml-app.vcxproj

@@ -104,7 +104,7 @@
       <SDLCheck>true</SDLCheck>
       <PreprocessorDefinitions>_DEBUG;_CONSOLE;%(PreprocessorDefinitions)</PreprocessorDefinitions>
       <ConformanceMode>Default</ConformanceMode>
-      <AdditionalIncludeDirectories>C:\mlpack\armadillo-11.4.1\include;C:\mlpack\mlpack-4.0.1\include\;C:\mlpack\cereal-1.3.2\include;C:\mlpack\ensmallen-2.19.0\include\%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
+      <AdditionalIncludeDirectories>C:\mlpack\armadillo-11.4.1\include;C:\mlpack\mlpack-4.1.0\include\;C:\mlpack\cereal-1.3.2\include;C:\mlpack\ensmallen-2.19.0\include\%(AdditionalIncludeDirectories)</AdditionalIncludeDirectories>
       <LanguageStandard>stdcpp17</LanguageStandard>
       <OpenMPSupport>false</OpenMPSupport>
       <AdditionalOptions>/Zc:__cplusplus %(AdditionalOptions)</AdditionalOptions>

doc/user/build_windows.md

@@ -261,5 +261,5 @@ If you are facing issues during the build process of mlpack, you may take a look
 at other third-party tutorials for Windows, but they may be out of date:
 
  * [Github wiki Windows Build page](https://github.com/mlpack/mlpack/wiki/WindowsBuild)
- * [Keon's tutorial for mlpack 2.0.3](http://keon.io/mlpack-on-windows)
+ * [Keon's tutorial for mlpack 2.0.3](https://keon.github.io/mlpack-on-windows/)
  * [Kirizaki's tutorial for mlpack 2](https://overdosedblog.wordpress.com/2016/08/15/once_again/)

doc/user/sample_ml_app.md

@@ -27,13 +27,13 @@ dependencies in Release Mode).
 - Under C/C++ > General > Additional Include Directories add:
 ```
  - C:\mlpack\armadillo-9.800.3\include
- - C:\mlpack\mlpack-4.0.1\src
+ - C:\mlpack\mlpack-4.1.0\src
  - C:\mlpack\ensmallen-2.19.0\include
  - C:\mlpack\cereal-3.1.2\include
 ```
 - Under Build Events > Post-Build Event > Command Line add:
 ```
- - xcopy /y "C:\mlpack\mlpack-4.0.1\packages\OpenBLAS.0.2.14.1\lib\native\bin\x64\*.dll" $(OutDir)
+ - xcopy /y "C:\mlpack\mlpack-4.1.0\packages\OpenBLAS.0.2.14.1\lib\native\bin\x64\*.dll" $(OutDir)
 ```
 
 *Note*: recent versions of Visual Studio set "Conformance Mode" enabled by

src/mlpack/core/util/version.hpp

@@ -17,8 +17,8 @@
 // The version of mlpack.  If this is a git repository, this will be a version
 // with higher number than the most recent release.
 #define MLPACK_VERSION_MAJOR 4
-#define MLPACK_VERSION_MINOR 0
-#define MLPACK_VERSION_PATCH 2
+#define MLPACK_VERSION_MINOR 1
+#define MLPACK_VERSION_PATCH 1
 
 // The name of the version (for use by --version).
 namespace mlpack {

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));
+  #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 network itself.
+    ar(CEREAL_NVP(network));
+    ar(CEREAL_NVP(parameters));
 
-  // Serialize the expected input size.
-  ar(CEREAL_NVP(inputDimensions));
+    // 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();
+    // 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();
+      networkOutput.clear();
+      networkDelta.clear();
 
-    layerMemoryIsSet = false;
-    inputDimensionsAreSet = false;
+      layerMemoryIsSet = false;
+      inputDimensionsAreSet = false;
 
-    // The weights in `parameters` will be correctly set for each layer in the
-    // first call to Forward().
-  }
+      // 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/init_rules/network_init.hpp

@@ -71,7 +71,7 @@ class NetworkInitialization
         // Initialize the layer with the specified parameter/weight
         // initialization rule.
         const size_t weight = network[i]->WeightSize();
-        arma::Mat<eT> tmp = arma::mat(parameters.memptr() + offset,
+        arma::Mat<eT> tmp = arma::Mat<eT>(parameters.memptr() + offset,
             weight, 1, false, false);
         initializeRule.Initialize(tmp, tmp.n_elem, 1);
 

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));
+  #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();
-  }
+    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/ann/layer/hard_tanh.cpp

@@ -2,8 +2,12 @@
  * @file tests/ann/layer/hard_tanh.cpp
  * @author Vaibhav Pathak
  *
- * Tests the hard_tanh layer
+ * Tests the hard_tanh layer.
  *
+ * mlpack is free software; you may redistribute it and/or modify it under the
+ * terms of the 3-clause BSD license.  You should have received a copy of the
+ * 3-clause BSD license along with mlpack.  If not, see
+ * http://www.opensource.org/licenses/BSD-3-Clause for more information.
  */
 
 #include <mlpack/core.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()