Merge branch 'development' into tk/onnx-benchmark

Author: tkoskela

.github/workflows/ci.yml

@@ -124,6 +124,15 @@ jobs:
         # Fix bug in FFT3 (cf. https://github.com/FFTW/fftw3/issues/332)
         sed -i -e 's/^.*FFTW3LibraryDepends.cmake.*$//1' ${{github.workspace}}/local/lib*/cmake/*/FFTW3Config.cmake
 
+    - name: Install HDF5
+      run: |
+        wget https://support.hdfgroup.org/ftp/HDF5/releases/hdf5-1.14/hdf5-1.14.3/src/hdf5-1.14.3.tar.gz -O- | tar --no-same-owner -xz;
+        cd hdf5-1.14.3/
+        mkdir build && cd build
+        #CC=`which mpicc` CXX=`which mpicxx` ./configure --prefix=${{github.workspace}}/local --enable-parallel --enable-shared --enable-cxx
+        CC=`which mpicc` CXX=`which mpicxx` cmake .. -DBUILD_TESTING=OFF -DCMAKE_INSTALL_PREFIX=${{github.workspace}}/local -DHDF5_ENABLE_PARALLEL:BOOL=ON -DALLOW_UNSUPPORTED:BOOL=ON -DHDF5_BUILD_FORTRAN:BOOL=OFF -DHDF5_BUILD_CPP_LIB:BOOL=ON -DHDF5_ENABLE_SZIP_SUPPORT:BOOL=OFF
+        make -j2 install
+
     - name: Checkout SOPT
       uses: actions/checkout@v4
       with:
@@ -214,9 +223,10 @@ jobs:
       # Build your program with the given configuration
       run: |
         export CMAKE_PREFIX_PATH=${{github.workspace}}/local:$CMAKE_PREFIX_PATH
+        #export PATH=${{github.workspace}}/local:$PATH
         mkdir -p ${{github.workspace}}/build
         cd ${{github.workspace}}/build
-        cmake .. --fresh -DCMAKE_INSTALL_PREFIX=${{github.workspace}}/local -Donnxrt=ON -Ddocasa=OFF -Ddompi=${{matrix.mpi}} -Dopenmp=${{matrix.omp}} -Dtests=ON
+        cmake .. --fresh -DCMAKE_INSTALL_PREFIX=${{github.workspace}}/local -Donnxrt=ON -Dhdf5=ON -Ddocasa=OFF -Ddompi=${{matrix.mpi}} -Dopenmp=${{matrix.omp}} -Dtests=ON
         make -j$(nproc --ignore 1) install
 
     - name: Test
@@ -278,7 +288,7 @@ jobs:
         export CMAKE_PREFIX_PATH=${{github.workspace}}/local:$CMAKE_PREFIX_PATH
         mkdir -p ${{github.workspace}}/build
         cd ${{github.workspace}}/build
-        cmake .. --fresh -DCMAKE_INSTALL_PREFIX=${{github.workspace}}/local -Donnxrt=ON -Ddompi=OFF -Dopenmp=OFF -Ddocs=ON
+        cmake .. --fresh -DCMAKE_INSTALL_PREFIX=${{github.workspace}}/local -Donnxrt=ON -Dhdf5=ON -Ddompi=OFF -Dopenmp=OFF -Ddocs=ON
         make -j$(nproc --ignore 1) install
 
     - name: Deploy to GH pages

CMakeLists.txt

@@ -16,6 +16,7 @@ option(docasa         "Enable CASA"                                     off)
 option(docs           "Build documentation"                             off)
 option(coverage       "Build coverage"                                  off)
 option(onnxrt         "Build with ONNXruntime interface"                off)
+option(hdf5           "Build with HDF5 interface"                       off)
 
 if(NOT CMAKE_BUILD_TYPE)
   message(STATUS "Setting build type to 'Release' as none was specified.")

README.md

@@ -91,6 +91,34 @@ When `purify` runs a directory will be created, and the output images will be
 saved and time-stamped. Additionally, a config file with the settings used will
 be saved and time-stamped, helping for reproducibility and book-keeping.
 
+Uncertainty Quantification
+--------------------------
+
+Bayesian hypothesis testing may be performed with the `purify_UQ` application, which can be located in the `build` directory. 
+
+This application takes a config yaml file with the following parameters:
+- `confidence_interval` or `alpha`. (alpha = 1 - confidence interval.))
+- `measurements_path`: path to measurements data (.vis file)
+- `reference_image_path`: path to reference image i.e. output from purify run. (.fits files)
+- `surrogate_image_path`: path to surrogate image i.e. doctored image with blurring or structural change that you want to test. (.fits file)
+- `sigma`: standard deviation for Gaussian likelihood. 
+- `gamma`: multiplicative factor for prior.
+- `purify_config`: path to purify config used to generate the reference image. **This should be used if available in order to ensure consistency of things like measurement and wavelet operators.**
+
+You can then run the uncertainty quantification with the command:
+```
+purify_UQ <path to UQ_config yaml>
+```
+
+The application will report the value of the objective function for each image, the threshold value calculated from the reference image, and whether the surrogate image is ruled out or not. 
+
+Presently this is designed to work for the unconstrained problem where:
+- The negative log-likelihood is a scaled L2-norm i.e. $ \frac{1}{2 \sigma^2} \sum (y_i - \Phi x_i)$ for some data $y$, image $x$, and measurement operator $\Phi$. (Equivalent to indepdendent multivariate Gaussian likelihood.)
+- The negative log prior is a scaled L1-norm in _wavelet space_ i.e. $\gamma \sum (\Psi^\dag x)_i$ for some image $x$ and wavelet operator $\Psi$.
+- The objective function is the sum of these two terms.
+
+Docker
+-------
 ## Debugging the CI workflow with tmate
 
 The CI workflow has a manual dispatch trigger which allows you to log into the job while it's running. You can trigger it in 

cmake_files/LookUpHighFive.cmake

@@ -0,0 +1,40 @@
+# Installs BlueBrain/HighFive into build directory
+#
+# - GIT_REPOSITORY: defaults to origin BlueBrain/HighFive repo on github
+# - GIT_TAG: defaults to master
+include(ExternalProject)
+if(HighFive_ARGUMENTS)
+  cmake_parse_arguments(HighFive "" "GIT_REPOSITORY;GIT_TAG" ""
+    ${HighFive_ARGUMENTS})
+endif()
+if(NOT HighFive_GIT_REPOSITORY)
+  set(HighFive_GIT_REPOSITORY https://github.com/BlueBrain/HighFive)
+endif()
+if(NOT HighFive_GIT_TAG)
+  set(HighFive_GIT_TAG master)
+endif()
+
+set(HighFive_DIR "${CMAKE_INSTALL_PREFIX}/external")
+ExternalProject_Add(
+  Lookup-HighFive
+  GIT_REPOSITORY ${HighFive_GIT_REPOSITORY}
+  GIT_TAG ${HighFive_GIT_TAG}
+  PREFIX "${CMAKE_BINARY_DIR}/external"
+  #INSTALL_DIR ${EXTERNAL_ROOT}
+  CMAKE_ARGS
+  -DCMAKE_INSTALL_PREFIX=${HighFive_DIR}
+  -DCMAKE_INSTALL_LIBDIR=${CMAKE_SHARED_LIBRARY_PREFIX}
+  -DHIGHFIVE_EXAMPLES=OFF
+  -DHIGHFIVE_USE_BOOST=OFF
+  -DHIGHFIVE_UNIT_TESTS=OFF
+  # Wrap download, configure and build steps in a script to log output
+  UPDATE_COMMAND ""
+  LOG_DOWNLOAD ON
+  LOG_CONFIGURE ON
+  LOG_BUILD ON
+  LOG_INSTALL ON
+)
+set(HighFive_INCLUDE_DIR "${HighFive_DIR}/include")
+#set(HighFive_LIBRARY_DIR "${HighFive_DIR}/${CMAKE_SHARED_LIBRARY_PREFIX}")
+#set(HighFive_LIBRARIES "HighFive")
+

cmake_files/dependencies.cmake

@@ -92,7 +92,11 @@ find_package(fftw3 NAMES FFTW3 REQUIRED)
 set(PURIFY_MPI FALSE)
 if(dompi)
   find_package(MPI REQUIRED)
-  set(PURIFY_MPI TRUE)
+  if (${sopt_HAS_MPI})
+    set(PURIFY_MPI TRUE)
+  else()
+    message(FATAL_ERROR "SOPT built without MPI support")
+  endif()
 endif()
 find_package(TIFF REQUIRED)
 
@@ -120,5 +124,23 @@ if(docasa)
   set(PURIFY_CASACORE TRUE)
 endif()
 
+set(PURIFY_H5 FALSE)
+if(hdf5)
+  find_package(HDF5 COMPONENTS CXX REQUIRED)
+  message(STATUS "Found HDF5 include dir: ${HDF5_INCLUDE_DIR}")
+  message(STATUS "Found HDF5 CXX library: ${HDF5_CXX_LIBRARIES}")
+  message(STATUS "HDF5 parallel: ${HDF5_IS_PARALLEL}")
+  if (PURIFY_MPI AND NOT ${HDF5_IS_PARALLEL})
+    message(FATAL_ERROR "HDF5 built without MPI support")
+  endif()
+  find_package(HighFive QUIET)
+  if(NOT HighFive_FOUND)
+    message(STATUS "HighFive not found. Attempt to install...")
+    include(LookUpHighFive)
+    set(PURIFY_HIGHFIVE_LOOKUP TRUE)
+  endif()
+  set(PURIFY_H5 TRUE)
+endif()
+
 # Add script to execute to make sure libraries in the build tree can be found
 add_to_ld_path("${EXTERNAL_ROOT}/lib")

cpp/CMakeLists.txt

@@ -23,6 +23,8 @@ add_include_dir(
   ${yaml-cpp_INCLUDE_DIR}
   ${Cubature_INCLUDE_DIR}
   ${CImg_INCLUDE_DIR}
+  ${HDF5_INCLUDE_DIR}
+  ${HighFive_INCLUDE_DIR}
 )
 
 
@@ -45,6 +47,8 @@ if (docs)
   add_subdirectory(docs)
 endif()
 
+add_subdirectory(uncertainty_quantification)
+
 add_executable(purify main.cc)
 set_target_properties(purify PROPERTIES RUNTIME_OUTPUT_DIRECTORY ${PROJECT_BINARY_DIR})
 target_link_libraries(purify libpurify ${sopt_LIBRARIES})

cpp/benchmarks/algorithms.cc

@@ -115,7 +115,7 @@ BENCHMARK_DEFINE_F(AlgoFixture, ForwardBackwardOnnx)(benchmark::State &state) {
   m_fb = factory::fb_factory<sopt::algorithm::ImagingForwardBackward<t_complex>>(
       factory::algo_distribution::serial, m_measurements_transform, wavelets, m_uv_data, m_sigma,
       beta, gamma, m_imsizey, m_imsizex, m_sara.size(), state.range(3) + 1, true, true, false, 1e-3,
-      1e-2, 50, 1.0, tf_model_path, factory::g_proximal_type::TFGProximal);
+      1e-2, 50, 1.0, tf_model_path, nondiff_func_type::Denoiser);
 
   while (state.KeepRunning()) {
     auto start = std::chrono::high_resolution_clock::now();

cpp/benchmarks/algorithms_mpi.cc

@@ -188,7 +188,7 @@ BENCHMARK_DEFINE_F(AlgoFixtureMPI, FbOnnxDistributeImage)(benchmark::State &stat
   m_fb = factory::fb_factory<sopt::algorithm::ImagingForwardBackward<t_complex>>(
       factory::algo_distribution::mpi_serial, m_measurements_distribute_image, wavelets, m_uv_data,
       m_sigma, beta, gamma, m_imsizey, m_imsizex, m_sara.size(), state.range(3), true, true,
-      false, 1e-3, 1e-2, 50, 1.0, tf_model_path, factory::g_proximal_type::TFGProximal);
+      false, 1e-3, 1e-2, 50, 1.0, tf_model_path, nondiff_func_type::Denoiser);
 
   // Benchmark the application of the algorithm
   while (state.KeepRunning()) {