v1.27.0

DiffEqFlux v1.27.0

Diff since v1.26.0

Merged pull requests:

• leanify the dependencies (#454) (@ChrisRackauckas)

v1.26.0

DiffEqFlux v1.26.0

Diff since v1.25.0

Merged pull requests:

• CompatHelper: bump compat for "StaticArrays" to "1.0" (#449) (@github-actions[bot])
• MassInstallAction: Install the GitHubActionsCI workflow on this repository (#450) (@christopher-dG)

v1.25.0

DiffEqFlux v1.25.0

Diff since v1.24.0

Closed issues:

• How to calculate loss in a given time budget? (#434)
• dt <= dtmin Aborting. There is either an error in your model specification or the true solution. (#435)
• DiffEqFlux Failed to Train with Error "type BinaryHeap has no filed comparer" (#441)

Merged pull requests:

• Add warning to sciml_train, add a dispatch of sciml_train to GalacticOptim and rename layer_sciml test file to galacticoptim (#424) (@Vaibhavdixit02)
• Another strategy to avoid local minima (#431) (@ghost)
• MassInstallAction: Install the TagBot workflow on this repository (#442) (@christopher-dG)
• add an example with exogenous input (#444) (@ChrisRackauckas)
• ready a release and see test results (#447) (@ChrisRackauckas)

v1.24.0

DiffEqFlux v1.24.0

Diff since v1.23.0

Closed issues:

• Training of UDEs with recurrent networks (#391)
• Question regarding solution type (#418)
• NEURAL ODE example with GPU not working (#419)

Merged pull requests:

• Work around missing matmul dispatch in CUDA.jl (#420) (@ChrisRackauckas)
• Add GalacticOptim and update tests to use it (#421) (@Vaibhavdixit02)
• CompatHelper: add new compat entry for "GalacticOptim" at version "0.2" (#423) (@github-actions[bot])
• update GalacticOptim calls (#428) (@ChrisRackauckas)
• CompatHelper: bump compat for "Distributions" to "0.24" (#430) (@github-actions[bot])

v1.23.0

DiffEqFlux v1.23.0

Diff since v1.22.0

Closed issues:

• FFJORD for batched data (#342)
• derivative definition of ude (#360)
• Optimizing over a neural network and a (time-independent) coefficient simultaneously (#409)
• CNF layers very slow with moderately sized Neural Networks (#413)

Merged pull requests:

• MNIST example DataLoader syntax bugfix (#369) (@mbeltagy)
• Update collocation.jl (#406) (@ChrisRackauckas)
• Update TagBot.yml (#410) (@kanav99)
• Batched Implementation for FFJORD (#415) (@avik-pal)
• work around trace for neldermead (#416) (@ChrisRackauckas)
• Reduce allocations in collocation (#417) (@baggepinnen)

v1.22.0

DiffEqFlux v1.22.0

Diff since v1.21.0

Closed issues:

• DDE Example leads to " StackOverflowError:" (#396)
• Differentiating with ReverseDiff fails on complex (#397)
• Fail to compile (Julia 1.5.1) (#399)
• sciml_train with Universal ODE fails on GPU (#400)

Merged pull requests:

• add hybrid example (#395) (@ChrisRackauckas)
• Splat the MNIST Data for proper batching (#398) (@avik-pal)
• Setup GPU tests to use CUDA (#404) (@ChrisRackauckas)

v1.21.0

DiffEqFlux v1.21.0

Diff since v1.20.0

Closed issues:

• Direct access of DESolution fields during Zygote differentiation (#321)
• EnsembleGPUArray() fails (#381)
• example notebook not working (#388)
• Handling sol.u interface (#389)
• KernelError: passing and using non-bitstype argument (#390)

Merged pull requests:

• CompatHelper: add new compat entry for "LoggingExtras" at version "0.4" (#392) (@github-actions[bot])
• CompatHelper: add new compat entry for "ConsoleProgressMonitor" at version "0.1" (#393) (@github-actions[bot])
• CompatHelper: add new compat entry for "TerminalLoggers" at version "0.1" (#394) (@github-actions[bot])

v1.20.0

DiffEqFlux v1.20.0

Diff since v1.19.0

v1.19.0

DiffEqFlux v1.19.0

Diff since v1.18.0

Closed issues:

• HamiltonianProblem not supported (#53)
• Hamiltonian Graph Networks with ODE Integrators (#80)
• torchdiffeq comparison (#322)
• Hamiltonian Neural Networks (#357)
• Error on the "Optimization of Ordinary Differential Equations" Tutorial (#376)
• ERROR: MethodError: no method matching -(::NTuple{29,Float64}) (#383)

Merged pull requests:

• Add support for Hamiltonian Neural Networks (#370) (@avik-pal)
• Regularized ffjord [WIP] (#372) (@d-netto)
• docs (#382) (@d-netto)
• additional fixes in cnf documentation (#384) (@d-netto)

v1.18.0

DiffEqFlux v1.18.0

Diff since v1.17.0

Closed issues:

• Flesh out documentation (#244)
• Multiple ODE solves from a single time series on GPU (#279)
• Zygote.gradient doesn't work with basic solve example (#344)
• Is it possible to learn neuralode for stiff problem of Robertson's problem? (#367)
• More pictures (#375)

Merged pull requests:

• spline: more tests (#345) (@d-netto)
• Spline docs (#348) (@d-netto)
• Finish GDE PR (#355) (@ChrisRackauckas)
• Use travis_wait (#356) (@christopher-dG)
• add smoothed collocation fitting (#359) (@ChrisRackauckas)
• change in ffjord tests (#361) (@d-netto)
• Update ffjord.jl (#363) (@nirmal-suthar)
• revised minibatch docs (#365) (@collinwarner)
• CompatHelper: add new compat entry for "MLDataUtils" at version "0.5" (#366) (@github-actions[bot])
• Require Zygote 0.5 (#373) (@ChrisRackauckas)
• test the fixed ensemble parallelism (#374) (@ChrisRackauckas)