v1.3.2

DiffEqFlux v1.3.2

Diff since v1.3.1

Closed issues:

• Feed known DE in neural DE (#122)
• Using loss as a function of data with sciml_train (#144)
• Some animations (#146)
• Box Constrained Optimization with sciml_train (#148)
• solve and concrete_solve crash when called through sciml_train (#153)
• ODE_layer (#155)
• FastDense is not found (#156)
• Documentation of sciml_train does not match actual arg prototype (#162)

Merged pull requests:

• Install TagBot as a GitHub Action (#145) (@JuliaTagBot)
• Add table of contents to README (#147) (@ali-ramadhan)
• [WIP] Add box constrained optimization dispatch in sciml_train (#151) (@abhigupta768)
• CompatHelper: bump compat for "Juno" to "0.8" (#157) (@github-actions[bot])
• Add an optional data argument and fix docstring (#163) (@ChrisRackauckas)
• Fix data splat in Flux optimizer version of sciml_train (#165) (@ChrisRackauckas)

v1.3.1

v1.3.1 (2020-02-07)

Diff since v1.3.0

Closed issues:

• No method matching finite\_difference\_jacobian (#142)
• modify parameters in-place in sciml_train! (#140)
• Fix StaticDense (#134)
• Expose more controls in sciml_train! (#132)

Merged pull requests:

• Remove UnsafeArrays (#143) (YingboMa)
• fast adjoints for fast layers (#141) (ChrisRackauckas)
• fix a bunch of GPU tests (#137) (ChrisRackauckas)

v1.3.0

v1.3.0 (2020-02-06)

Diff since v1.2.2

Closed issues:

• TrackerAdjoint on GPUs (#135)

Merged pull requests:

• stopping criterion on current loss (#139) (ric-cioffi)

v1.2.2

v1.2.2 (2020-02-04)

Diff since v1.2.1

Merged pull requests:

• add some abstract types (#136) (ChrisRackauckas)

v1.2.1

v1.2.1 (2020-02-02)

Diff since v1.2.0

v1.2.0

v1.2.0 (2020-02-02)

Diff since v1.1.1

Closed issues:

• NeuralODE not training with DiffEqFlux v1.1.1 (#127)
• Better SciML Training Loop (#120)
• DiffEqFlux could handle infinite loss better. (#70)

Merged pull requests:

• CompatHelper: add new compat entry for "UnsafeArrays" at version "1.0" (#131) (github-actions[bot])
• CompatHelper: add new compat entry for "StaticArrays" at version "0.12" (#130) (github-actions[bot])
• CompatHelper: add new compat entry for "Optim" at version "0.20" (#129) (github-actions[bot])
• CompatHelper: add new compat entry for "Juno" at version "0.7" (#128) (github-actions[bot])
• implement fast versions of Flux (#126) (ChrisRackauckas)
• Introduce sciml_train (#125) (ChrisRackauckas)

v1.1.1

v1.1.1 (2020-01-24)

Diff since v1.1.0

Closed issues:

• Suggestion: Passing Args to Callback in Flux.train!, and return args from Loss() (#116)

Merged pull requests:

• patch Zygote GPU support (#117) (ChrisRackauckas)

v1.1.0

v1.1.0 (2020-01-20)

Diff since v1.0.0

Closed issues:

• Neural SDE example has convergence and plotting issue (#114)
• Tests sometimes fail (#112)
• LoadError: UndefVarError: Tracker not defined (#110)
• Stack Overflow error (#88)
• DiffEqFlux 0.6.0, 0.5.0 and 0.5.1 do not compile on julia 1.0 (#81)

Merged pull requests:

• NeuralDSDE, NeuralSDE, and NeuralCDDE (#115) (ChrisRackauckas)

v1.0.0

v1.0.0 (2020-01-13)

Diff since v0.10.2

Closed issues:

• Long integrations fail with Zygote even with hardcoded adjoint (#108)
• Cannot provide higher level functions with neural networks with Zygote: Neural ODE restructure dimension mismatch (#107)
• Long integrations fail with Zygote even with hardcoded adjoint (#105)
• Zygote GPU (#104)
• Training a neural_ode with Flux is VERY slow on the GPU. (#102)
• Remaining Zygote issues (#101)
• DiffEqFlux v0.4.0 will not pre-compile with Flux 0.10.0 in Julia 1.3 (#99)
• Zygote compatibility for new Flux update (#79)

Merged pull requests:

• Neural DE struct layers (#109) (ChrisRackauckas)
• Neural ODE via restructure/destructure (#103) (ChrisRackauckas)
• [WIP] Refactoring destructure (#97) (dhairyagandhi96)
• Zygote Compatibility (#96) (ChrisRackauckas)

v0.10.2

v0.10.2 (2019-12-29)

Diff since v0.10.1