v1.41.0

DiffEqFlux v1.41.0

Diff since v1.40.1

Merged pull requests:

• CompatHelper: bump compat for "GalacticOptim" to "2" (#585) (@github-actions[bot])

v1.40.1

DiffEqFlux v1.40.1

Diff since v1.40.0

Closed issues:

• Precompilation fails: UndefVarError: HasBranchingCtx not defined (#583)

v1.40.0

DiffEqFlux v1.40.0

Diff since v1.39.0

Closed issues:

• Multiple shooting example doesn't work (#564)
• Calling cpu on FastChain fails. (#578)

Merged pull requests:

• add hasbranching overload to auto-optimize (#580) (@ChrisRackauckas)
• Use return codes in multiple shooting (#581) (@adrhill)
• add device(x::FastLayer) warning (#582) (@ChrisRackauckas)

v1.39.0

DiffEqFlux v1.39.0

Diff since v1.38.0

Closed issues:

• Compiling Tuple{typeof(task_local_storage) try/catch is not supported. (#561)
• Parameter Estimation Example not converged in a low error (#569)
• Typo in minibatching example (#570)

Merged pull requests:

• Fix #570 (#572) (@avik-pal)
• Simplify and improve sciml_train heuristics (#574) (@ChrisRackauckas)
• CompatHelper: add new compat entry for "Reexport" at version "1" (#575) (@github-actions[bot])
• CompatHelper: add new compat entry for "Optim" at version "1" (#576) (@github-actions[bot])

v1.38.0

DiffEqFlux v1.38.0

Diff since v1.37.0

Closed issues:

• How to create a multilayer perceptron using DiffEqFlux.jl (#554)

Merged pull requests:

• Revert "Update divergence.md" (#559) (@ChrisRackauckas)
• fix typo (#562) (@IlyaOrson)
• Update bouncing_ball.md ReverseDiffAdjoint example (#565) (@Vaibhavdixit02)

v1.37.0

DiffEqFlux v1.37.0

Diff since v1.36.1

Closed issues:

• Probably user error: sciML_train func running but not optimizing (#528)
• StackOverflow Error when Backproping through SteadyStateProblem (#531)
• DiffEqFlux.sciml_train fails with save_idxs keyword (#534)
• Regarding Neural ODE Documentation (#538)
• Speed and innacuracy of sciml_train with BFGS() on previously demonstrated code (#539)
• FFJORD for flexible batched data (#540)
• MethodError in sequential learning problem (#543)
• Discrete Callback causes Error: immutable struct of type FakeIntegrator cannot be changed (#550)
• No method matching similar(::Tuple{...}, ::Int64) (#552)
• Mass matrix example requires mass to always equal 1 (#553)
• False gradient for FastDense when bias=false (#556)

Merged pull requests:

• fix typo (#530) (@spin0r)
• Convolutional Neural ODE for MNIST Classification Demo (#532) (@avik-pal)
• Fix typos in docs (#533) (@adrhill)
• Fix & simplify multiple shooting (#522) (#535) (@adrhill)
• Update copy paste example without NeuralODE constructor (#541) (@Vaibhavdixit02)
• Generalize continuity loss in multiple shooting (#542) (@adrhill)
• Fix grammar (#545) (@namangup)
• Add solver kwargs to multiple shooting (#546) (@adrhill)
• CompatHelper: bump compat for "Distributions" to "0.25" (#547) (@github-actions[bot])
• Add FAQ on divergent trajectories (#548) (@ChrisRackauckas)
• Update divergence.md (#549) (@Vaibhavdixit02)
• Fix and test no bias adjoint (#558) (@ChrisRackauckas)

v1.36.1

DiffEqFlux v1.36.1

Diff since v1.36.0

v1.36.0

DiffEqFlux v1.36.0

Diff since v1.35.1

Closed issues:

• Discretize-Optimize vs Optimize-Discretize (#463)
• error in precompiling DiffEqFlux v1.35.0 (#518)
• How to treat parameters in a model data struture? (#520)

Merged pull requests:

• Optimize the Neural SDE Demo (#519) (@avik-pal)
• Use other solvers than Tsit5 in multiple_shoot (#521) (@adrhill)
• Adapt is good now? (#525) (@ChrisRackauckas)

v1.35.1

DiffEqFlux v1.35.1

Diff since v1.35.0

v1.35.0

DiffEqFlux v1.35.0

Diff since v1.34.1

Closed issues:

• Update Neural SDE documentation (#402)
• Neural DAE tests not included in runtests and outdated (#425)
• Do Multiple Shooting (#481)
• FastDense error on GPU (#501)
• Make docs nicer (#504)
• ODE-LSTMs (#508)
• Is there any difference in neuralODE and ODENet? (#510)
• Error with non-in-place vector ODEs (#512)
• sciml_train summary is gone? (#513)
• Error in the Copy-Paste Code for Optimization of Ordinary Differential Equations (#517)

Merged pull requests:

• Updated sciml_train information. (#500) (@mkg33)
• Do not free memory you do not own. (#503) (@aussetg)
• Do multiple shooting (#505) (@Manas2030)
• Update Flux.md (#506) (@yewalenikhil65)
• Restructure documentation (#507) (@ChrisRackauckas)
• CompatHelper: bump compat for "Flux" to "0.12" (#514) (@github-actions[bot])
• fix typo (#515) (@ArnoStrouwen)
• Disable graph stuff to get tests running again (#516) (@ChrisRackauckas)