🥮 mooncakes.io
    Sign in

    autodiff

    Forward-mode automatic differentiation over Luna Flow algebraic and arithmetic structures.

    autodiff
    dual-numbers
    math
    Download zip
    Author
    Version
    0.3.1
    License
    Apache-2.0
    Repository
    Last updated
    4 hours ago
    Downloads
    37

    Dependencies

    #autodiff

    Luna-Flow/autodiff provides forward-mode automatic differentiation for MoonBit. Run a function on dual numbers with and you get its value and its exact first derivative in one pass, with no step size and no symbolic algebra. Dual[T] is generic over the Luna Flow scalar traits, and bridges compute gradients and Jacobians over linear-algebra vectors and derivatives of luna-poly polynomials.

    #Install

    moon add Luna-Flow/autodiff@0.3.1

    // moon.pkg
    import {
      "Luna-Flow/autodiff",
    }

    #Example

    fn main {
      let (value, slope) = @autodiff.value_and_diff(x => x * x + x.sin(), 2.0)
      println("f(2) = \{value}, f'(2) = \{slope}")
    }

    f(2) = 4.909297426825682, f'(2) = 3.5838531634528574

    #Packages

    PackagePurpose
    Luna-Flow/autodifffacade: Dual, diff, value_and_diff, re-exported traits and error types
    Luna-Flow/autodiff/dualDual[T]: arithmetic, derivative rules, checked division and square root, trait instances
    Luna-Flow/autodiff/forwardscalar drivers diff and value_and_diff
    Luna-Flow/autodiff/corealgebraic facade (Dual and luna-generic structure traits)
    Luna-Flow/autodiff/elementaryanalytic facade (Sqrt, Exponential, Logarithmic, Trigonometric, Constants)
    Luna-Flow/autodiff/checkedchecked facade (DivChecked, SqrtChecked, ArithmeticContext, ArithmeticError)
    Luna-Flow/autodiff/linalggradient, jacobian and their value_and_* forms over linear-algebra/immut
    Luna-Flow/autodiff/polyderivatives of dense and univariate sparse luna-poly polynomials

    Dual[T] implements the ring-level traits of luna-generic and no Field, Inverse or order, because has no inverse. Gradients use one forward pass per input. Reverse mode, higher-order (jet) types, symbolic differentiation and multivariate polynomial partial derivatives are not part of this release.

    #Toolchain

    MoonBit moonc 0.10 or newer, with moon.mod / moon.pkg manifests. Tests run on wasm-gc, wasm, js and native.

    The checked operations return the errors of the scalar's own instances; the checked API lists them for Double from Luna-Flow/arithmetic 0.5.

    #Documentation

    The manual, with API, tutorial and design pages for every package and Chinese and Japanese translations, is published at https://lunaflow.cn/en/autodiff/. Its English source is doc/manual/index.md.

    #Contributing

    See CONTRIBUTING.md and the correctness checklist. Run ./ready_to_pr.sh before opening a pull request.

    #License

    Apache-2.0, as declared in moon.mod.

    AddGroup

    source

    AddMonoid

    source

    ArithmeticContext

    source

    ArithmeticError

    source

    ArithmeticErrorKind

    source

    Constants

    source

    DivChecked

    source

    Dual

    source

    Exponential

    source

    FromInteger

    source

    The canonical map out of the integers: the unique ring homomorphism ℤ -> Self. It must agree with from_natural on non-negative arguments.

    FromNat

    source

    The canonical map out of the naturals: the unique semiring homomorphism ℕ -> Self. The argument must be non-negative.

    Logarithmic

    source

    MulMonoid

    source

    One

    source

    Ring

    source

    RoundingMode

    source

    Semiring

    source

    Sqrt

    source
    using @Luna-Flow/arithmetic { trait Sqrt }

    SqrtChecked

    source

    Trigonometric

    source

    Zero

    source

    diff

    source

    value_and_diff

    source

    Source Files

    Powered by MoonBit

    Site sourceReport issuePackagesBuild queueSkillsStatistics

    © 2026 mooncakes.io