Probnum

New Features
- Base class `StoppingCriterion` for probabilistic numerical methods which supports boolean arithmetic (561, 570)
- Bayesian quadrature can be run in an iterative fashion (420)

Improvements
- Efficient arithmetic for linear operators (505)

Bug Fixes
- Optimized memory consumption of `Constant` random variables (551, 553)
- Matrix-variate normal matrix multiplication now works for all possible cases (558, 572)

Refactoring
- Belief updates for probabilistic linear solvers (546)
- Package structure improved (560, 568)
- Removed support for complex `dtypes` of `LinearOperators` (575)

Testing
- Optional arguments can now be passed to `pytest` via `tox` (563)

Other
- Dropped support for Python 3.6 (548)

New Features
- Kernel broadcasting (539): Allows efficient evaluation of kernels without forming a full kernel matrix.

Improvements
- Updated quickstart guides for linear solvers (511) and linear operators (534)

Refactoring
- Probabilistic linear solvers: Information operators (494) and stopping criteria (499)

Documentation
- Corrected documentation of some ODE example problems (544)

Improvements
- New Lorenz systems as ODE example problems (515)
- Improved package build process (78cd1ed9e8e3b1302aae59d736ce1ccc4d61cbc3, 993a27524e91420253e2096dec1d845d02456ebf)

Refactoring
- Improvements to the SDE API (501)
- Removal of global `pylint` disables (522)
- Restructured `.gitignore` (525)

Documentation
- Fixes to tutorials and Sphinx build (516)

New Features
- Event handling for ODE solvers (498)

Improvements
- Interface function for perturbation-based ODE solvers (481)
- Use of `LinearOperators` in `statespace.IBM` (479)

Refactoring
- Reorganization of `diffeq` subpackage (477)
- Restructuring of `filtsmooth` subpackage (478)
- Refactoring of `statespace` subpackage (492)
- Components (state, belief, policy) of probabilistic linear solvers as separate classes (460)

Documentation
- Automatically generated documentation for `config` options (480)
- Various updates to tutorials (495, 506, 510)

New Features
- Perturbed step ODE solvers (382)
- Global configuration module to set package-wide defaults (465)

Improvements
- Random sparse matrices are now generated in a memory efficient way (455, 457)
- On-the-fly construction of filtering posteriors (403)

Refactoring
- Renamed type module to `probnum.typing` (454)
- Removal of `random_state` attribute from random variables (387)
- ODE examples moved into problem zoo (475)

Documentation
- New website and logo (433)

New Features
- Abstract base class `ProbabilisticNumericalMethod` 410

Refactoring
- Simplification of `RandomVariable` instantiation code 425

Bug Fixes
- Fixed tests for extended Kalman filters 409
- `LinearOperator.inv()` no longer throws warnings for sparse matrices 434

Documentation
- updated tutorials 419 423
- restructured contribution guide 414