Amici

Features/enhancements:
- Add computation of residuals, residuals sensitivity, Fisher information matrix (223)
- More efficient conversion of std::vector to numpy ndarray (375)
- Allow specifying timepoints in ExpData (370)

Minor fixes:
- Condition parameters in ExpData now only temporarily override Model parameters (371)
- Ensure non-negative states for Newton solver (378)

Features/enhancements:
- Check SBML model validity (343)
- Allow per-parameter setting of amioptions::pscale from matlab interface (350)
- Documentation

Major fixes:
- Don't compile main.cpp into python model module which broke modules if amici was compiled without libhdf5 (363)

Minor fixes:
- Fix compiler warnings (353)
- Plotting, SBML example mode, ...

Features:
- Added symbol names to python-wrapped models and make available via Model.getParameterNames(), model.getStateNames(), ...
- Extended Python interface example

Python package available via pypi: https://pypi.org/project/amici/0.7.3/

Features:
- Python package: more flexible HDF5 library localization
- Extended CI: python tests, preequilibration tests, run in venv

Major bugfixes:
- Fix python sbml model import / compilation error (undefined function)
- Fix model preequilibration

Minor fixes:
- Various fixes for mingw compilation of python source distribution
- Cmake compatibility with < 3.8 restored

Features:
- Allow specifying sigma-parameters from Python interface

Major bugfixes:
- Fix dsigma_y/dp and downstream sensitivity errors

- Major revision of documentation
- Improved Python interface
- More comprehensive Python interface example
- Fixed sensitivity computation in Python-generated models
- Various other bug fixes

WARNING:
- For models with sigma-parameters and dsigma_y/dp != 0, dsigma_y/dp was computed incorrectly. This propagates to all dependent sensitivities. This applies also to some older releases and has been fixed in v0.7.1.