Releases: JuliaDecisionFocusedLearning/ImplicitDifferentiation.jl
Releases · JuliaDecisionFocusedLearning/ImplicitDifferentiation.jl
v0.9.1
v0.9.0
ImplicitDifferentiation v0.9.0
Breaking changes
- Switch to KrylovKit for linear solve
- Remove options from
IterativeLinearSolverandOperatorRepresentation - Remove preparation-related options from the
ImplicitFunctionconstructor
Merged pull requests:
- test: use ChainRulesTestUtils (#175) (@gdalle)
- Revert "test: use ChainRulesTestUtils" (#177) (@gdalle)
- refactor!: split out preparation (#179) (@gdalle)
- revamp!: use KrylovKit for type flexibility (beyond Vector), split out preparation (#180) (@gdalle)
- fix: only factorize for direct linear solve (#181) (@gdalle)
v0.8.1
ImplicitDifferentiation v0.8.1
Merged pull requests:
v0.8.0
ImplicitDifferentiation v0.8.0
Breaking changes
- Renamed
KrylovLinearSolverintoIterativeLinearSolver - Strengthened typing in iterative solve, which will make ComponentArrays error unless the
OperatorRepresentationis switched from LinearOperators to LinearMaps - Split the
backendkeyword toImplicitFunctionintobackends(one forxand one fory) - Activated
strictmode with DifferentiationInterface to catch typing errors
Merged pull requests:
v0.7.2
v0.7.1
What's Changed
- Update ForwardDiff.jl compat by @longemen3000 in #164
- Bump version by @gdalle in #165
New Contributors
- @longemen3000 made their first contribution in #164
Full Changelog: v0.7.0...v0.7.1
v0.7.0
What's Changed
- Complete revamp by @gdalle in #160
- Remove mention of (now unsupported) kwargs by @gdalle in #162
- Delete build directory by @gdalle in #163
Main breaking changes
- The
forwardfunction (now calledsolver) must return a byproductzin all cases, which will usually benothing. It used to be optional. For most users, adaptation will only require replacingforward(x) = ybyforward(x) = (y, nothing). - Support for keyword arguments has been removed.
- Forward rules for Enzyme have been dropped (temporarily).
Please refer to the API reference and FAQ for more details.
Main new features
- The ability to run preparation on the differentiation of
conditions. This is useful e.g. to leverage a sparse backend and build Jacobian matrices instead of calling lazy pushforwards and pullbacks
Full Changelog: v0.6.3...v0.7.0