v1.46.0

DiffEqFlux v1.46.0

Diff since v1.45.3

Closed issues:

• Enzyme segfault when using FastChain and AutoZygote (#691)
• DimensionMismatch: arrays could not be broadcast to a common size; (#701)

Merged pull requests:

• test Flux 0.13 (#699) (@ChrisRackauckas)

v1.45.3

DiffEqFlux v1.45.3

Diff since v1.45.2

Merged pull requests:

• add prediction-error method tutorial (#688) (@baggepinnen)
• Disable Graph Conv test (#693) (@ChrisRackauckas)

v1.45.2

DiffEqFlux v1.45.2

Diff since v1.45.1

Merged pull requests:

• No compats on test deps (#692) (@ChrisRackauckas)

v1.45.1

DiffEqFlux v1.45.1

Diff since v1.45.0

Merged pull requests:

• typo fixes for hybrid NODE examples (#683) (@frankschae)
• Added Scalar dispatch for FastDense (#686) (@Abhishek-1Bhatt)

v1.45.0

DiffEqFlux v1.45.0

Diff since v1.44.1

Closed issues:

• Use faster tanh? (#674)
• saveat type not matching time type can result in missed tstops in values not exactly represented in floating point numbers (#675)
• AutoZygote seems to result in segfaults (#676)

Merged pull requests:

• Preallocation Option for vjp calculation (#671) (@Abhishek-1Bhatt)
• Update and restructuring of docs on autojacvec options (#678) (@JordiBolibar)
• Use NNlib fast activations (#679) (@ChrisRackauckas)
• CompatHelper: add new compat entry for NNlib at version 0.8, (keep existing compat) (#682) (@github-actions[bot])

v1.44.1

DiffEqFlux v1.44.1

Diff since v1.44.0

Closed issues:

• Exposing DiffEqParamEstim's Tooling (#121)
• WARNING: could not import DistributionsAD._mv_categorical_logpdf into ReverseDiffX (#649)
• Wrong Jacobian Matrix (#650)
• Reproducibility on different operating systems (#655)
• ERROR: Compiling ...: try/catch is not supported. (#660)
• Error: Version input must not be null (#663)
• Scalar indexing during GPU training in example (#668)
• [sciml_train] ArgumentError: Initial x is not in (despite clamping io) (#669)

Merged pull requests:

• use zero matrix each time in jacobian_fn (#651) (@prbzrg)
• Update BayesianNODE_NUTS.md (#654) (@QiyaoWei)
• update default AD choice docs (#657) (@hexaeder)
• Fix CI version choices (#664) (@ChrisRackauckas)
• Fix for various changes (#666) (@ChrisRackauckas)
• unmark fixed GPU test (#667) (@ChrisRackauckas)
• Fix default optimizer when parameter bounds are given (#670) (@ChrisRackauckas)

v1.44.0

DiffEqFlux v1.44.0

Diff since v1.43.0

Closed issues:

• Regularized FFJORD (#343)
• Mixed CPU+GPU adjoints (#401)
• ReverseDiff/DiffEqSensitivity, BoundsError: attempt to access 56-element Array{Float54,1} at index [0] (#563)
• Steady State GPU Training Hangs and Quits (#567)
• maxiters: only works with optimization algs in sciml_train? (#598)
• MethodError: no method matching has_syms(::SciMLBase.OptimizationFunction) (#600)
• zeros32 not defined (#606)
• GPU example throws errors (#608)
• ERROR: LoadError: Cannot determine ordering of Dual tags ForwardDiff.Tag (#610)
• Troubles with running the simple example code on: https://diffeqflux.sciml.ai/dev/ (#616)
• Julia crashes during parameter estimation for a system of differential equations (#619)
• Migrate to new style test-setup (#626)
• Error when running copy-pasteable code from stable documentation (#634)
• Optimization Issue in FFJORD with AutoZygote (#635)
• Errors running unit tests (#644)
• Slower performance using the GPU (#645)
• tests failing with MethodError: _axes(::StrideArraysCore (#646)

Merged pull requests:

• small typo (#611) (@mahdiarsadeghi)
• Update note on sensitivity analysis for hybrid DEs (#613) (@frankschae)
• use DiagNormal instead of FullNormal (#617) (@prbzrg)
• code cleaning for ffjord (#620) (@prbzrg)
• add smoke test for ffjord (#621) (@prbzrg)
• Update other tests for FFJORD (#625) (@prbzrg)
• add FFJORD sampler (#627) (@prbzrg)
• Update References (#629) (@prbzrg)
• fix a typo (#631) (@prbzrg)
• Remove type promotion in FFJORD (#633) (@prbzrg)
• fix wrong variable in neural_ode_sciml.md (#637) (@wkirgsn)
• update Project.toml file (#642) (@prbzrg)
• Update Geometric Flux API (#643) (@avik-pal)

v1.43.0

DiffEqFlux v1.43.0

Diff since v1.42.1

Closed issues:

• multiple_shoot for EnsembleProblem (#601)

Merged pull requests:

• multiple_shoot for EnsembleProblem (#602) (@vboussange)
• CompatHelper: add new compat entry for SciMLBase at version
  1 , (keep existing compat) (#605) (@github-actions[bot])

v1.42.1

DiffEqFlux v1.42.1

Diff since v1.42.0

Closed issues:

• :nothing initialised in FastChain (#593)
• Zygote sensitivities not working for ensemble ODE (#595)

Merged pull requests:

• Update divergence.md (#551) (@Vaibhavdixit02)
• Fix "dense.W" deprecation warning (#596) (@prbzrg)
• some cosmetic changes (#597) (@prbzrg)
• Fix differences between full code and documented code in CNF example (#599) (@prbzrg)

v1.42.0

DiffEqFlux v1.42.0

Diff since v1.41.0

Closed issues:

• Possible Indexing error in multiple_shoot (#522)
• incorrect initialise params in FastDense after update to Flux v0.12.5 (#594)

Merged pull requests:

• Remove unfinished sentence in sciml_train docs (#589) (@gileshd)
• Tutorial edits (#590) (@Gregliest)
• use Flux.zeros32 (#592) (@DhairyaLGandhi)