Computational reproducibility

前置知识

40 项必需的前置知识（一直追溯到底）

你将学会

• Package a computation with pinned dependencies, tests and a single command that reproduces its results.
• Classify a computational claim as exact, numerical, statistical or experimental, and state its limitations.
• Archive code and data with persistent identifiers, and cite software properly.

定义

• A computation is reproducible if others can obtain the same results from the same code, data and environment.
• A pinned environment fixes the exact versions of all software dependencies.
• Exact computation (integer or symbolic arithmetic) differs from floating-point approximation, whose results need error bounds to be meaningful.

定理

• Floating-point results can differ between hardware and library versions, so tolerances must be stated.
• A computer-assisted proof depends on the correctness of its code; formal verification or independent reimplementation strengthens the claim.

常见误解

• Thinking a notebook that ran once is reproducible.
• Reporting agreement to many digits as if it were proof.

可选发现

• Formalising published mathematics Formal verification is the strongest form of reproducibility for proofs, including computer-assisted ones.
• ← Statistical inference

学习路径

• Research preparation

延伸阅读

• Nicholas J. Higham, Accuracy and Stability of Numerical Algorithms (2002)
• Geir Kjetil Sandve, Anton Nekrutenko, James Taylor and Eivind Hovig, Ten Simple Rules for Reproducible Computational Research (2013)