Adapt mlpack CI workflow to ensmallen (#461)

* Adapt mlpack CI workflow to ensmallen.

* Explicitly build the tests.

* Try to figure out what is going on with Bandicoot search on macOS.

* Remove unnecessary CMake options that aren't used by ensmallen.

* Turn on verbose build for debugging.

* Make sure OpenMP is found before compiling with it.

* We have to find the threads library before compiling with it.

* Maybe this is the safer approach?

* A little extra safety.

* Reorganization and get some more output.

* Only use OpenMP for the tests if it was totally found.

* Hopefully fix names of CMake variables for Windows.

* Clean up debugging output.

* Try to debug the one failing test; I can't reproduce locally.

* Oops, order the test running correctly to prevent early exits.

* Improve output of NSGA-2.

* See what it does on the regular matrix type too.

* Try and get some more output.

* Oops, fix the ordering.

* More stuff in the wrong order.

* Another small error.

* Fix incorrect child calculation in NSGA-2 (thanks @zoq for finding this).

* Clean up after debugging.

* Update to action version v1.2.23 to avoid Node version warning.

* Merge adjacent blocks.

* Update changelog.

* Let's just try three cores and see what happens... maybe it's different for ensmallen...

* One more entry in the changelog.

* Enable ccache if it's found.

* Use more concise notation.

.appveyor.yml

@@ -1,59 +0,0 @@
-environment:
-  ARMADILLO_DOWNLOAD: "https://sourceforge.net/projects/arma/files/armadillo-10.8.2.tar.xz"
-  BLAS_LIBRARY: "%APPVEYOR_BUILD_FOLDER%/OpenBLAS.0.2.14.1/lib/native/lib/x64/libopenblas.dll.a"
-  BLAS_LIBRARY_DLL: "%APPVEYOR_BUILD_FOLDER%/OpenBLAS.0.2.14.1/lib/native/lib/x64/libopenblas.dll"
-
-  matrix:
-    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017
-      VSVER: Visual Studio 15 2017 Win64
-      MSBUILD: C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\MSBuild\15.0\Bin\MSBuild.exe
-
-    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2019
-      VSVER: Visual Studio 16 2019
-      MSBUILD: C:\Program Files (x86)\Microsoft Visual Studio\2019\Community\MSBuild\Current\Bin\MSBuild.exe
-
-    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2022
-      VSVER: Visual Studio 17 2022
-      MSBUILD: C:\Program Files\Microsoft Visual Studio\2022\Community\MSBuild\Current\Bin\MSBuild.exe
-
-configuration: Release
-
-install:
-  - ps: nuget install OpenBLAS -o "${env:APPVEYOR_BUILD_FOLDER}"
-
-build_script:
-  # First, download and build Armadillo.
-  - cd ..
-  - appveyor DownloadFile %ARMADILLO_DOWNLOAD% -FileName armadillo.tar.xz
-  - 7z x armadillo.tar.xz -so -txz | 7z x -si -ttar > nul
-  - cd armadillo-10.8.2 && mkdir build && cd build
-  - >
-    cmake -G "%VSVER%"
-    -DBLAS_LIBRARY:FILEPATH=%BLAS_LIBRARY%
-    -DLAPACK_LIBRARY:FILEPATH=%BLAS_LIBRARY%
-    -DCMAKE_PREFIX:FILEPATH="%APPVEYOR_BUILD_FOLDER%/armadillo"
-    -DBUILD_SHARED_LIBS=OFF
-    -DCMAKE_POLICY_VERSION_MINIMUM=3.5
-    -DCMAKE_BUILD_TYPE=Release ..
-  - >
-    "%MSBUILD%" "armadillo.sln"
-    /m /verbosity:quiet /p:Configuration=Release;Platform=x64
-  - cd ../..
-
-  # Now build ensmallen.
-  - cd ensmallen && mkdir build && cd build
-  - >
-    cmake -G "%VSVER%"
-    -DARMADILLO_INCLUDE_DIR=%APPVEYOR_BUILD_FOLDER%/../armadillo-10.8.2/include/
-    -DARMADILLO_LIBRARIES=%BLAS_LIBRARY%
-    -DLAPACK_LIBRARY=%BLAS_LIBRARY%
-    -DBLAS_LIBRARY=%BLAS_LIBRARY%
-    -DCMAKE_POLICY_VERSION_MINIMUM=3.5
-    -DCMAKE_BUILD_TYPE=Release ..
-  - >
-    "%MSBUILD%" "ensmallen.sln"
-    /m /verbosity:minimal /nologo /p:BuildInParallel=false
-
-  # Run tests after copying libraries.
-  - ps: cp C:\projects\ensmallen\OpenBLAS.0.2.14.1\lib\native\bin\x64\*.* C:\projects\ensmallen\build\
-  - ctest -C Release -V --output-on-failure .

.github/workflows/ci.yml

@@ -0,0 +1,230 @@
+# ci.yml: defines all the CI jobs run for each commit and pull request.
+
+name: Continuous Integration
+
+on:
+  workflow_dispatch:
+  push:
+    branches: [master, main]
+  pull_request:
+    branches: [master, main]
+  release:
+    types: [published, created, edited]
+permissions:
+  contents: read
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  ###
+  ### Linux and macOS test matrix for all binding types.
+  ###
+  ci_matrix:
+    strategy:
+      fail-fast: false # Run all configurations even if one fails.
+      matrix:
+        config: [
+          {
+            runner: ubuntu-latest,
+            os: 'Linux',
+            name: 'Debug',
+            cmakeVars: '-DCMAKE_BUILD_TYPE=Debug',
+            cores: 4
+          },
+          {
+            runner: ubuntu-latest,
+            os: 'Linux',
+            name: 'Release',
+            cmakeVars: '-DCMAKE_BUILD_TYPE=Release',
+            cores: 4
+          },
+          {
+            runner: macOS-latest,
+            os: 'macOS',
+            name: 'Debug',
+            cmakeVars: '-DCMAKE_BUILD_TYPE=Debug',
+            cores: 3
+          },
+          {
+            runner: macOS-latest,
+            os: 'macOS',
+            name: 'Release',
+            cmakeVars: '-DCMAKE_BUILD_TYPE=Release',
+            cores: 3
+          }
+        ]
+
+    name: '${{ matrix.config.os }} (${{ matrix.config.name }})'
+    if: ${{ github.repository == 'mlpack/ensmallen' }}
+    runs-on: ${{ matrix.config.runner }}
+
+    steps:
+      - uses: actions/checkout@v6
+
+      # Set up ccache.
+      - name: Get ccache
+        uses: hendrikmuhs/ccache-action@v1.2.23
+        with:
+          key: ${{ matrix.config.os }}-${{ matrix.config.name }}
+          append-timestamp: false
+          variant: ccache
+          max-size: 2G # Github has a 10GB repo cache limit, so we use 2GB/job.
+
+      - name: Configure ccache
+        shell: bash
+        run: |
+          ccache --set-config "sloppiness=include_file_ctime"
+          ccache --set-config "hash_dir=false"
+          ccache --zero-stats
+
+      # Install build dependencies via apt.
+      - name: Install build dependencies (Linux)
+        if: matrix.config.os == 'Linux'
+        run: |
+          sudo apt-get update
+          sudo apt-get install --yes --allow-unauthenticated libopenblas-dev
+          # Install the oldest Armadillo version that we support.
+          curl -k -L https://sourceforge.net/projects/arma/files/armadillo-10.8.2.tar.xz | tar -xvJ
+          cd armadillo*
+          # Armadillo 10.8.2 requests CMake 2.8.12+ compatibility, but that has
+          # been dropped from newer CMake versions.  However, with Armadillo's
+          # CMake configuration, it works just fine to specify
+          # CMAKE_POLICY_VERSION_MINIMUM to ignore the missing reverse
+          # compatibility.
+          cmake -DCMAKE_POLICY_VERSION_MINIMUM=3.5 .
+          make
+          sudo make install
+          cd ..
+
+      - name: Install build dependencies (macOS)
+        if: matrix.config.os == 'macOS'
+        run: |
+          sudo xcode-select --switch /Applications/Xcode.app/Contents/Developer
+          brew install libomp openblas armadillo ccache
+
+      - name: Configure ensmallen with CMake
+        run: |
+          mkdir build && cd build
+          cmake ${{ matrix.config.cmakeVars }} ../
+
+      - name: Build ensmallen tests
+        run: cd build && make -j${{ matrix.config.cores }} ensmallen_tests
+
+      # Run the tests manually so that we can get JUnit output.
+      - name: Run ensmallen_tests
+        run: |
+          cd build
+          # The use of only one thread is to prevent thrashing on older versions
+          # of OpenBLAS (0.3.26 and older) where OpenMP and pthreads aren't
+          # playing together well.
+          OMP_NUM_THREADS=1 ./ensmallen_tests -r junit -o ensmallen_tests.junit.xml
+
+      - name: Parse test output
+        uses: rcurtin/test-summary-action@dist
+        if: success() || failure()
+        with:
+          paths: "build/*.junit.xml"
+          show: "fail, skip"
+          fail_job: true
+          print_output: true
+
+  ###
+  ### Windows CI testing.
+  ###
+
+  ci_windows:
+    name: 'Windows build: VS2022'
+    if: ${{ github.repository == 'mlpack/ensmallen' }}
+    runs-on: windows-2022
+
+    steps:
+      - uses: actions/checkout@v6
+
+      # Set up ccache.
+      - name: Get ccache
+        uses: hendrikmuhs/ccache-action@v1.2.23
+        with:
+          key: ${{ runner.os }}
+          append-timestamp: false
+          variant: ccache
+          max-size: 2G
+
+      # Set up Visual Studio.
+      - name: Set up Visual Studio
+        uses: microsoft/setup-msbuild@v2
+
+      # Set up paths for MS build tools.
+      - name: Set up devcmd
+        uses: ilammy/msvc-dev-cmd@v1
+        with:
+          arch: x64
+
+      # Install build dependencies.
+      - name: Install build dependencies (Windows)
+        shell: powershell
+        run: |
+          nuget install OpenBLAS -o deps
+
+      # Install the oldest supported version of Armadillo and install cereal.
+      - name: Install some dependencies manually
+        shell: bash
+        run: |
+          cd deps/
+          curl -L https://sourceforge.net/projects/arma/files/armadillo-10.8.2.tar.xz -o armadillo-10.8.2.tar.xz
+          tar -xf armadillo-10.8.2.tar.xz
+          cd armadillo-10.8.2/
+          cmake -G "Ninja" \
+            -DCMAKE_BUILD_TYPE=Release \
+            -DCMAKE_POLICY_VERSION_MINIMUM=3.5 \
+            -DCMAKE_C_COMPILER=cl \
+            -DCMAKE_CXX_COMPILER=cl \
+            -DBLAS_LIBRARY:FILEPATH=${{ github.workspace }}/deps/OpenBLAS.0.2.14.1/lib/native/lib/x64/libopenblas.dll.a \
+            -DLAPACK_LIBRARY:FILEPATH=${{ github.workspace }}/deps/OpenBLAS.0.2.14.1/lib/native/lib/x64/libopenblas.dll.a \
+            -DBUILD_SHARED_LIBS=OFF .
+          # Disable the wrapper because transitive linking doesn't work on
+          # Windows anyway.
+          cp tmp/include/armadillo_bits/config.hpp tmp/include/armadillo_bits/config.hpp.tmp
+          cat tmp/include/armadillo_bits/config.hpp.tmp | sed 's|#define ARMA_USE_WRAPPER|// #define ARMA_USE_WRAPPER|' \
+              > tmp/include/armadillo_bits/config.hpp
+
+      - name: Configure ensmallen with CMake
+        shell: bash
+        run: |
+          mkdir build && cd build
+          cmake -G "Ninja" \
+              -DCMAKE_BUILD_TYPE=Release \
+              -DCMAKE_C_COMPILER=cl \
+              -DCMAKE_CXX_COMPILER=cl \
+              -DARMADILLO_INCLUDE_DIR=../deps/armadillo-10.8.2/tmp/include/ \
+              -DBLAS_LIBRARY='${{ github.workspace }}'/deps/OpenBLAS.0.2.14.1/lib/native/lib/x64/libopenblas.dll.a \
+              -DLAPACK_LIBRARY='${{ github.workspace }}'/deps/OpenBLAS.0.2.14.1/lib/native/lib/x64/libopenblas.dll.a \
+              -DCMAKE_VERBOSE_MAKEFILE=ON \
+              ../
+
+      - name: Build mlpack
+        run: cd build && cmake --build . --config Release --target ensmallen_tests
+
+      # Run the tests manually so that we can get JUnit output.
+      - name: Run ensmallen_tests
+        shell: bash
+        run: |
+          # The .dlls are stored in the bin/ directory, and those are the ones
+          # we need to run with.
+          cp -v '${{ github.workspace }}'/deps/OpenBLAS.0.2.14.1/lib/native/bin/x64/* build/
+          ./ensmallen_tests.exe -r junit > ensmallen_tests.junit.xml || true
+          # Remove version numbers and other non-xml from the output.
+          cat ensmallen_tests.junit.xml | sed '/<?xml/,$!d' > ensmallen_tests.junit.xml.tmp
+          rm ensmallen_tests.junit.xml
+          mv ensmallen_tests.junit.xml.tmp ensmallen_tests.junit.xml
+
+      - name: Parse test output
+        uses: rcurtin/test-summary-action@dist
+        if: success() || failure()
+        with:
+          paths: "build/*.junit.xml"
+          show: "fail, skip"
+          fail_job: true
+          print_output: true
+

.travis.yml

@@ -1,39 +0,0 @@
-os: linux
-dist: focal
-language: cpp
-
-env:
-  - ARMADILLO=latest SANITY_HISTORY=perform
-  - ARMADILLO=minimum
-
-stages:
-  - test
-  - name: sanity
-    if: type = pull_request AND env(SANITY_HISTORY) = "perform"
-
-jobs:
-  include:
-    - stage: sanity
-      name: "HISTORY.md Check"
-      script: sh ./scripts/history-update-check.sh
-
-script:
-  - sudo apt-get update
-  - sudo apt-get install -y --allow-unauthenticated libopenblas-dev liblapack-dev g++ xz-utils
-  - if [ $ARMADILLO == "latest" ]; then
-      curl https://ftp.fau.de/macports/distfiles/armadillo/`curl https://ftp.fau.de/macports/distfiles/armadillo/ -- | grep '.tar.xz' | sed 's/^.*<a href="\(armadillo-[0-9]*.[0-9]*.[0-9]*.tar.xz\)".*$/\1/' | tail -1` | tar xvJ && cd armadillo*;
-    else
-      curl -L https://sourceforge.net/projects/arma/files/armadillo-10.8.2.tar.xz | tar -xvJ && cd armadillo*;
-    fi
-  - cmake . && make && sudo make install && cd ..
-  - mkdir build && cd build && cmake .. && make ensmallen_tests -j2
-  - CTEST_OUTPUT_ON_FAILURE=1 travis_wait 30 ctest -j2
-
-notifications:
-  email:
-    - mlpack-git@lists.mlpack.org
-  irc:
-    channels:
-      - "chat.freenode.net#mlpack"
-    on_success: change
-    on_failure: always

CMake/FindBandicoot.cmake

@@ -142,10 +142,10 @@ if(EXISTS "${BANDICOOT_INCLUDE_DIR}/bandicoot_bits/config.hpp")
         find_package(CUDAToolkit REQUIRED)
 
         if (CUDAToolkit_FOUND)
-	  message(STATUS "CUDA includes: ${CUDAToolkit_INCLUDE_DIRS}")
-	  message(STATUS "CUDA libraries: ${CUDAToolkit_LIBRARY_DIR}")
+          message(STATUS "CUDA includes: ${CUDAToolkit_INCLUDE_DIRS}")
+          message(STATUS "CUDA libraries: ${CUDAToolkit_LIBRARY_DIR}")
 
-	  set(CUDA_LIBRARIES CUDA::cudart CUDA::cuda_driver)
+          set(CUDA_LIBRARIES CUDA::cudart CUDA::cuda_driver)
           set(CUDA_CUBLAS_LIBRARIES CUDA::cublas)
           set(CUDA_curand_LIBRARY CUDA::curand)
           set(CUDA_cusolver_LIBRARY CUDA::cusolver)
@@ -155,10 +155,10 @@ if(EXISTS "${BANDICOOT_INCLUDE_DIR}/bandicoot_bits/config.hpp")
               "${CUDAToolkit_INCLUDE_DIRS}")
           set(SUPPORT_LIBRARIES "${SUPPORT_LIBRARIES}"
               CUDA_LIBRARIES
-	      CUDA_CUBLAS_LIBRARIES
-	      CUDA_curand_LIBRARY
-	      CUDA_cusolver_LIBRARY
-	      CUDA_nvrtc_LIBRARY)
+              CUDA_CUBLAS_LIBRARIES
+              CUDA_curand_LIBRARY
+              CUDA_cusolver_LIBRARY
+              CUDA_nvrtc_LIBRARY)
           set(HAVE_CUDA true)
         endif()
       endif ()

CMakeLists.txt

@@ -43,6 +43,12 @@ target_include_directories(ensmallen INTERFACE
   $<BUILD_INTERFACE:${PROJECT_SOURCE_DIR}/include>
   $<INSTALL_INTERFACE:include>)
 
+# Use ccache if we find it.
+find_program(CCACHE_PROGRAM ccache)
+if(CCACHE_PROGRAM)
+  set_property(GLOBAL PROPERTY RULE_LAUNCH_COMPILE "${CCACHE_PROGRAM}")
+endif()
+
 # Set warning flags for target.
 if(MSVC)
   target_compile_options(ensmallen INTERFACE $<BUILD_INTERFACE:/Wall>)
@@ -52,16 +58,25 @@ endif()
 
 # Find OpenMP and link it.
 if(USE_OPENMP)
-  if(NOT TARGET OpenMP::OpenMP_CXX)
-    find_package(Threads REQUIRED)
-    add_library(OpenMP::OpenMP_CXX IMPORTED INTERFACE)
-    set_property(TARGET OpenMP::OpenMP_CXX
-        PROPERTY INTERFACE_COMPILE_OPTIONS ${OpenMP_CXX_FLAGS})
-    # Only works if the same flag is passed to the linker; use CMake 3.9+ otherwise (Intel, AppleClang).
-    set_property(TARGET OpenMP::OpenMP_CXX
-        PROPERTY INTERFACE_LINK_LIBRARIES ${OpenMP_CXX_FLAGS} Threads::Threads)
-  endif()
-  target_link_libraries(ensmallen INTERFACE OpenMP::OpenMP_CXX)
+  find_package(OpenMP)
+  if (OpenMP_FOUND)
+    if(NOT TARGET OpenMP::OpenMP_CXX)
+      find_package(Threads REQUIRED)
+      if (Threads_FOUND)
+        add_library(OpenMP::OpenMP_CXX IMPORTED INTERFACE)
+        set_property(TARGET OpenMP::OpenMP_CXX
+            PROPERTY INTERFACE_COMPILE_OPTIONS ${OpenMP_CXX_FLAGS})
+        # Only works if the same flag is passed to the linker; use CMake 3.9+ otherwise (Intel, AppleClang).
+        set_property(TARGET OpenMP::OpenMP_CXX
+            PROPERTY INTERFACE_LINK_LIBRARIES ${OpenMP_CXX_FLAGS} Threads::Threads)
+      endif ()
+    endif ()
+
+    if (TARGET OpenMP::OpenMP_CXX)
+      target_compile_definitions(ensmallen INTERFACE -DENS_USE_OPENMP)
+      target_link_libraries(ensmallen INTERFACE OpenMP::OpenMP_CXX)
+    endif ()
+  endif ()
 endif()
 
 if(USE_BANDICOOT)

CONTRIBUTING.md

@@ -13,6 +13,23 @@ can happen either here on Github or in the `#mlpack:matrix.org` channel on
 [here](https://www.mlpack.org/doc/developer/community.html) for more
 information.
 
+## Usage of LLMs / Coding Assistants
+
+Each contributor needs to understand the content they are proposing for ensmallen,
+and how the proposed content changes and/or expands functionality in ensmallen.
+Content includes source code, documentation, and other material such as images
+and datasets.  The pull request must justify the proposed content in the
+description.  Using coding assistants or tools like Large Language Models (LLMs) 
+does not grant additional privileges or reduce our expectations.
+
+Each contributor is responsible for the proposed content, regardless of where
+the content came from.  The responsibility includes ensuring that the content
+can be submitted to ensmallen and does not violate intellectual property rights
+such as copyright(s).  Source code in ensmallen is licensed under the BSD license;
+see `LICENSE.txt` for details.
+
+## Pull request process
+
 Once a pull request is submitted, it must be reviewed and approved before a
 merge, to ensure that:
 
@@ -109,8 +126,8 @@ $ cd ensmallen
 
 # - or -
 
-$ wget http://ensmallen.org/files/ensmallen-3.11.0.tar.gz
-$ tar -xvzpf ensmallen-3.11.0.tar.gz
+$ wget http://ensmallen.org/files/ensmallen-3.11.1.tar.gz
+$ tar -xvzpf ensmallen-3.11.1.tar.gz
 $ cd ensmallen-latest
 ```
 

HISTORY.md

@@ -1,5 +1,18 @@
 ### ensmallen ?.??.?: "???"
 ###### ????-??-??
+ * Fix incorrect child generation computation for NSGA-II
+   ([#461](https://github.com/mlpack/ensmallen/pull/461)).
+
+ * Better handling of OpenMP/pthreads linking for non-Linux systems
+   ([#461](https://github.com/mlpack/ensmallen/pull/461)).
+
+### ensmallen 3.11.1: "Sunny Day"
+###### 2026-07-25
+ * Loosen tolerance for Lookahead tests to prevent spurious failures
+   ([#456](https://github.com/mlpack/ensmallen/pull/456)).
+
+ * Fix uninitialized gradient constraint matrix for `AugLagrangian`
+   ([#458](https://github.com/mlpack/ensmallen/pull/458)).
 
 ### ensmallen 3.11.0: "Sunny Day"
 ###### 2025-12-15

include/ensmallen_bits/aug_lagrangian/aug_lagrangian_function_impl.hpp

@@ -73,7 +73,8 @@ void AugLagrangianFunction<LagrangianFunction, VecType>::Gradient(
   gradient.zeros();
   function.Gradient(coordinates, gradient);
 
-  GradType constraintGradient; // Temporary for constraint gradients.
+  // Temporary for constraint gradients.
+  GradType constraintGradient(gradient.n_rows, gradient.n_cols);
   for (size_t i = 0; i < function.NumConstraints(); i++)
   {
     function.GradientConstraint(i, coordinates, constraintGradient);

include/ensmallen_bits/config.hpp

@@ -48,6 +48,7 @@
   #define ENS_PRAGMA_OMP_ATOMIC   _Pragma("omp atomic")
   #define ENS_PRAGMA_OMP_CRITICAL _Pragma("omp critical")
   #define ENS_PRAGMA_OMP_CRITICAL_NAMED _Pragma("omp critical(section)")
+  #include <omp.h>
 #else
   #define ENS_PRAGMA_OMP_PARALLEL
   #define ENS_PRAGMA_OMP_ATOMIC

include/ensmallen_bits/ens_version.hpp

@@ -16,16 +16,16 @@
 // The minor version is two digits so regular numerical comparisons of versions
 // work right.  The first minor version of a release is always 10.
 #define ENS_VERSION_MINOR 11
-#define ENS_VERSION_PATCH 0
+#define ENS_VERSION_PATCH 1
 // If this is a release candidate, it will be reflected in the version name
 // (i.e. the version name will be "RC1", "RC2", etc.).  Otherwise the version
 // name will typically be a seemingly arbitrary set of words that does not
 // contain the capitalized string "RC".
 #define ENS_VERSION_NAME "Sunny Day"
 // Incorporate the date the version was released.
-#define ENS_VERSION_YEAR "2025"
-#define ENS_VERSION_MONTH "12"
-#define ENS_VERSION_DAY "15"
+#define ENS_VERSION_YEAR "2026"
+#define ENS_VERSION_MONTH "07"
+#define ENS_VERSION_DAY "25"
 
 namespace ens {
 

include/ensmallen_bits/nsga2/nsga2_impl.hpp

@@ -168,7 +168,14 @@ typename MatType::elem_type NSGA2::Optimize(
   for (size_t generation = 1; generation <= maxGenerations && !terminate;
       generation++)
   {
-    Info << "NSGA2: iteration " << generation << "." << std::endl;
+    Info << "NSGA2: iteration " << generation << "; objectives [";
+    if (numObjectives > 0)
+    {
+      Info << calculatedObjectives[0];
+      for (size_t i = 0; i < calculatedObjectives.size(); ++i)
+        Info << ", " << calculatedObjectives[i] << ", ";
+    }
+    Info << "]." << std::endl;
 
     // Create new population of candidate from the present elite population.
     // Have P_t, generate G_t using P_t.
@@ -347,7 +354,7 @@ void NSGA2::Crossover(
   // Use traits from parentA for indices where idx is 1 and parentB otherwise.
   childA = parentA % idx + parentB % (1 - idx);
   // Use traits from parentB for indices where idx is 1 and parentA otherwise.
-  childB = parentA % (1 - idx) + parentA % idx;
+  childB = parentA % (1 - idx) + parentB % idx;
 }
 
 //! Perform mutation of the candidates weights with some noise.

include/ensmallen_bits/problems/aug_lagrangian_test_functions_impl.hpp

@@ -54,8 +54,6 @@ inline void AugLagrangianTestFunction<MatType>::Gradient(
 {
   // f'_x1(x) = 12 x_1 + 4 x_2
   // f'_x2(x) = 4 x_1 + 6 x_2
-  gradient.set_size(2, 1);
-
   gradient[0] = 12 * coordinates[0] + 4 * coordinates[1];
   gradient[1] = 4 * coordinates[0] + 6 * coordinates[1];
 }
@@ -127,8 +125,6 @@ inline void GockenbachFunction::Gradient(const MatType& coordinates,
   // f'_x1(x) = 2 (x_1 - 1)
   // f'_x2(x) = 4 (x_2 + 2)
   // f'_x3(x) = 6 (x_3 + 3)
-  gradient.set_size(3, 1);
-
   gradient[0] = 2 * (coordinates[0] - 1);
   gradient[1] = 4 * (coordinates[1] + 2);
   gradient[2] = 6 * (coordinates[2] + 3);
@@ -166,8 +162,6 @@ inline void GockenbachFunction::GradientConstraint(const size_t index,
                                                    const MatType& coordinates,
                                                    GradType& gradient)
 {
-  gradient.zeros(3, 1);
-
   switch (index)
   {
     case 0:
@@ -184,6 +178,7 @@ inline void GockenbachFunction::GradientConstraint(const size_t index,
       // h'_x2(x) = 0
       // h'_x3(x) = 1
       gradient[0] = -2 * coordinates[0];
+      gradient[1] = 0;
       gradient[2] = 1;
       break;
   }

tests/CMakeLists.txt

@@ -65,6 +65,9 @@ set(ENSMALLEN_TESTS_SOURCES
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR})
 add_executable(ensmallen_tests EXCLUDE_FROM_ALL ${ENSMALLEN_TESTS_SOURCES})
 target_link_libraries(ensmallen_tests PRIVATE ensmallen)
+if (USE_OPENMP AND TARGET OpenMP::OpenMP_CXX)
+  target_link_libraries(ensmallen_tests PRIVATE OpenMP::OpenMP_CXX)
+endif ()
 
 # Copy test data into place.
 add_custom_command(TARGET ensmallen_tests

tests/lookahead_test.cpp

@@ -60,5 +60,6 @@ TEMPLATE_TEST_CASE("LookaheadAdam_LogisticRegressionFunction", "[Lookahead]",
   LogisticRegressionFunctionTest<TestType>(
       optimizer,
       Tolerances<TestType>::LRTrainAcc,
-      Tolerances<TestType>::LRTestAcc);
+      Tolerances<TestType>::LRTestAcc,
+      3 /* allow a couple trials */);
 }

tests/moead_test.cpp

@@ -560,8 +560,6 @@ TEMPLATE_TEST_CASE("DefaultMOEAD_ZDT1Function", "[MOEAD]", ENS_TEST_TYPES)
  */
 TEMPLATE_TEST_CASE("DirichletMOEAD_ZDT3Function", "[MOEAD]", ENS_TEST_TYPES)
 {
-  typedef typename TestType::elem_type ElemType;
-
   //! Parameters taken from original ZDT Paper.
   ZDT3<TestType> zdt3(300);
   const double lowerBound = 0;
@@ -591,4 +589,4 @@ TEMPLATE_TEST_CASE("DirichletMOEAD_ZDT3Function", "[MOEAD]", ENS_TEST_TYPES)
   opt.Optimize(objectives, coords, finalPopulation, finalFront);
 
   REQUIRE(VariableBoundsCheck(finalPopulation));
-}
+}

tests/nsga2_test.cpp

@@ -275,4 +275,4 @@ TEMPLATE_TEST_CASE("NSGA2_ZDTONEFunction", "[NSGA2]", ENS_ALL_CPU_TEST_TYPES)
   double g = 1. + 9. * sum / (static_cast<double>(numVariables - 1));
 
   REQUIRE(g == Approx(1.0).margin(0.99));
-}
+}

tests/parallel_sgd_test.cpp

@@ -45,8 +45,8 @@ TEMPLATE_TEST_CASE("ParallelSGDTest_SparseFunction", "[ParallelSGD]",
 
     size_t batchSize = std::ceil((float) f.NumFunctions() / i);
 
-    ConstantStep decayPolicy(0.4 * batchSize);
-    ParallelSGD<ConstantStep> s(10000, batchSize, 1e-5, true, decayPolicy);
+    ConstantStep decayPolicy(0.02 * batchSize);
+    ParallelSGD<ConstantStep> s(10000, batchSize, 1e-7, true, decayPolicy);
     FunctionTest<SparseTestFunction>(s,
         Tolerances<TestType>::LargeObj,
         Tolerances<TestType>::LargeCoord);
@@ -62,10 +62,10 @@ TEMPLATE_TEST_CASE("ParallelSGD_GeneralizedRosenbrockFunction",
     // Create the generalized Rosenbrock function.
     GeneralizedRosenbrockFunction f(i);
 
-    ConstantStep decayPolicy(0.001 * f.NumFunctions());
+    ConstantStep decayPolicy(0.00005 * f.NumFunctions());
 
     ParallelSGD<ConstantStep> s(
-        100000, f.NumFunctions(), Tolerances<TestType>::Obj / 100, true,
+        1000000, f.NumFunctions(), Tolerances<TestType>::Obj / 100, true,
         decayPolicy);
 
     TestType coordinates = f.GetInitialPoint<TestType>();
@@ -73,11 +73,11 @@ TEMPLATE_TEST_CASE("ParallelSGD_GeneralizedRosenbrockFunction",
     omp_set_num_threads(1);
     double result = s.Optimize(f, coordinates);
 
-    REQUIRE(result == Approx(0.0).margin(Tolerances<TestType>::Obj));
+    REQUIRE(result == Approx(0.0).margin(100 * Tolerances<TestType>::Obj));
     for (size_t j = 0; j < i; ++j)
     {
       REQUIRE(coordinates(j) ==
-          Approx(1.0).epsilon(Tolerances<TestType>::Coord));
+          Approx(1.0).epsilon(100 * Tolerances<TestType>::Coord));
     }
   }
 }