1.0.0a19

Improved Python stub generation and documentation.

Full Changelog: 1.0.0a18...1.0.0a19

1.0.0a18

This release adds a casadi::external function loader that does not depend on the CasADi C++ library. This simplifies the installation and reduces the binary size.

The ALPAQA_WITH_CASADI option now enables this loader by default. To enable the full CasADi loader, which also supports other types of casadi::Function objects (e.g. symbolic expressions or serialized functions), use ALPAQA_WITH_EXTERNAL_CASADI (this requires CasADi to be installed).

Full Changelog: 1.0.0a17...1.0.0a18

1.0.0a17

• New dynamic loading API (#10)
• Add TypeErasedProblem::get_name
• Convex Newton direction provider
• Clang 18 support
• Parameter JSON export
• Optimized parameter getters/setters
• More checks in gradient-checker
• Better Python type hints MinimizationProblemDescription
• Add CMake and Ninja as Python dependencies
• Improved documentation

Full Changelog: 1.0.0a16...1.0.0a17

1.0.0a16

• DLProblem support on Windows
• Bash and Zsh completion for alpaqa-driver and solver options
• Support for complex matrices, prox of complex ℓ₁ norm
• Allow problems to link to OpenMP (with ALPAQA_WITH_OPENMP)
• Improvements to DLProblem, TypeErased and print_csv

1.0.0a15

• Improved documentation
• Eigen 3.4.0 support
• Support comments in CSV files, support reading integers
• Improved Conan support
• CMake component support
• Improved packaging
• Conan + CMake example project

Full Changelog: 1.0.0a14...1.0.0a15

1.0.0a14

• Added experimental MATLAB/MEX interface (https://kul-optec.github.io/alpaqa/1.0.0a14/Sphinx/reference/matlab-api.html)
• Support specifying parameters using JSON
• ASCII aliases for parameter names
• Driver reports more problem information
• Document benchmarking strategies
• Added FISTA solver
• Improved L-BFGS-B solver
• Expose FISTASolver and LBFGSBSolver in Python
• Improved Clang and MSVC compatibility
• Improved Conan support
• Upgrade CasADi to 3.6.4
• Various minor improvements bug fixes

Full Changelog: 1.0.0a13...1.0.0a14

1.0.0a13

• Support for non-owning TypeErased references
• ALPAQA_STANDALONE option that configures relative RPATHs
• Improved performance SparsityConverter
• Verify Hessian-vector products in gradient checker
• Improved documentation
• Updated Python examples
• Added alpaqa for NMPC presentation slides
• More solver progress information
• Support alternative CUTEst collections
• Windows support for alpaqa-driver (no DLProblem support yet)
• Support Python CasADi problems without pre-compilation
• Parallel compilation for CasADi problems
• Improved py::object problem interface

Full Changelog: 1.0.0a12...1.0.0a13

1.0.0a12

• New sparse matrix representation: get_xyz_num_nonzeros() → get_xyz_sparsity()
• Conversions between sparsity patterns
• Add ALPAQA_DEBUG_CHECKS_EIGEN CMake option to control Eigen macros
• Fix option name in Python examples
• Expose DL and CUTEst loaders in Python on Linux
• Expose Hessian and Jacobian matrices in Python

Full Changelog: 1.0.0a11...1.0.0a12

1.0.0a11

• Fixed alpaqa::prox for NuclearNorm and L1Norm
• Improved handling of L > L_max in PANOC and ZeroFPR
• Added alpaqa::prox_step example

Full Changelog: 1.0.0a10...1.0.0a11

1.0.0a10

• More consistent QUB backtrack across algorithms
• Add ALPAQA_WITH_OPENMP option to enable (limited) parallelization
• Improved CUTEst loader
• Link to CUTEst as shared library
• Add alpaqa::prox and alpaqa::prox_step
• Initial set of proximal operators in alpaqa::functions (Indicator of box, ℓ₁-norm, nuclear norm)
• Update Python dependencies
• Disable float32 Python bindings by default
• GCC 10 and Clang 14 support
• Improved documentation

Full Changelog: 1.0.0a9...1.0.0a10