Anonymous function creation is slow, and almost always necessary

未关闭
#873 9 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
5/5
预计耗时
一周以上
新手友好度
35/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
活跃
技术栈
julia
领域
backend

调研方向

Start by reading the ForwardDiff method signatures that accept functions and initial values; the issue does not name specific files or tests. Check how DifferentialEquations handles call-site parameters, then determine the scope and compatibility implications of adding a similar option. Done means ForwardDiff can accept parameters at the call site without relying on a closure that captures changing state.

由索引模型根据 Issue 内容生成。

描述

Suppose I have a function f(x::Vector, p::SomeData), which I want to compute the Jacobian of. A common way to do this is

p = generate_some_data()
jac = ForwardDiff(x -> f(x, p), x0)

which has very bad performance on every call because we are instantiating a new anonymous function every time, which triggers compilation. A more clever way of doing this is

partially_applied_f = x -> f(x, p)
#repeat the lines below as much as needed
p = generate_some_data()
jac = ForwardDiff(partially_applied_f, x0)

which introduces a weird invisible dependency of the calls of partially_applied_f on the current state of p. Would it be feasible to update method signatures in this package to accept parameters at the call site? Such that the following code is valid:

p = generate_some_data()
jac = ForwardDiff(f, x0, parameter = p)

DifferentialEquations already do something similar with their method calls. This would remove a lot of overhead and imprevisibility in my use cases.

主要语言
Julia
星标
1k
派生
165
平均合并
8 天 22 小时
30 天内合并 PR
3

环境准备

这个项目没有提供开发容器、Dockerfile 或贡献指南,环境需要你自己搭建:先看它的 README,通用步骤见我们的新手贡献指南。

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

JuliaDiff/ForwardDiff.jl 的其他 Issue

查看 JuliaDiff/ForwardDiff.jl 的全部 Issue

相似的 Issue

更多 Julia Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。