- Support for FEniCS 2019.1.0
- Added function taylor_to_dict, which automatically computes shape derivatives and hessians without user input
- Added support for shape optimization. Supports "naive" shape optimization, computing shape derivatives with the ufl function CoordinateDerivative, as well as more advanced optimization, where only the boundary mesh nodes are the design variables. Demo can be found in examples/stokes-shape-ad.
- Added support for FunctionAssigner
- Reintroduced tape visualisation to graphviz dot files
- Added support for KrylovSolver and PETScKrylovSolver
- Support for FEniCS 2018.1.0
- Merged taylor_test and taylor_test_multiple
- Use tensorflow for tape graph visualisation
- Add ROLSolver to pyadjoint optimization package
- Renamed ReducedFunctional.optimize to optimize_tape
- Now annotates function assignment of linear combinations
- create_overloaded_object is moved to pyadjoint, introducing the OverloadedType._ad_init_object method for the same purpose.
- Added UFLConstraint to dolfin-adjoint
- BlockVariable now has an attribute marked_in_path, which indicates if one needs to compute (adjoint/tlm/hessian) values for this block variable.
- Block.evaluate_adj and Block.evaluate_hessian now take a new bool argument markings, indicating if relevant block variables are marked.
- Added alternative Block methods for evaluating adjoint/tlm/hessian values, which hides some of the technical implementation details common for most evaluate_* Block methods.
- Add the class Placeholder to pyadjoint
- Add the method Control.tape_value for retrieving the value of an object on the tape
- Initial release