본문으로 건너뛰기
cccskills
무료GitHub에 공개

rseng-performance-profiling

Covers making research code faster with evidence: profiling before optimizing (py-spy and language-native profilers), interpreting hotspots, choosing optimizations by measured payoff, benchmark regression tracking with airspeed velocity (asv), and scaling measurements. Use when the user says their code is slow, asks to optimize or speed something up, wants benchmarks or performance regression tests, or before recommending rewrites, parallelism or GPUs on performance grounds.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

fdiblen/rseng-agent-skills스타 24(GitHub 24 + cccskills 0)

실행 점수 미검증

설치 방법 보기

포함된 파일(2)

  • SKILL.md5.0 KB
  • references.md383 B

SKILL.md(원문)

설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.

Performance profiling and optimization

The cardinal rule: never optimize unprofiled code. Intuition about where research code spends time is wrong often enough that acting on it wastes effort and adds complexity to the wrong places. The discipline is a loop: measure, find the dominant cost, fix only that, measure again, stop when it is fast enough for the science. "Fast enough" is a research judgment - a one-off analysis needs no tuning; a per-particle inner loop run for months justifies serious work.

Profile first

  • Whole-program view: a sampling profiler (py-spy for Python: py-spy top --pid on a live process or py-spy record for a flame graph) shows where time actually goes with negligible overhead and no code changes. Language-native equivalents: perf (compiled code), Rprof (R), @profile/BenchmarkTools (Julia).
  • Line/function view once a hotspot is known: cProfile+snakeviz, line-profiler, or the IDE's profiler.
  • Memory matters separately: allocation churn and swapping look like CPU slowness; profile memory when RSS grows or the machine swaps.
  • Profile REPRESENTATIVE inputs at meaningful scale - toy inputs have different hotspots than production ones.

Report findings as fractions ("68% of runtime in distance_matrix"), not feelings; the fraction bounds the possible speedup (Amdahl's law) and justifies - or kills - the optimization.

Optimize in payoff order

  1. Algorithmic: a better algorithm or data structure beats any micro-optimization; check complexity before code details.
  2. Do less: cache repeated computation, hoist work out of loops, short-circuit, avoid recomputing unchanged pipeline stages (rseng-workflows).
  3. Use optimized building blocks: vectorized NumPy/BLAS operations, compiled library routines - most scientific speedups come from replacing interpreted loops with library calls (rseng-language-guides).
  4. Compile the hotspot: Numba/Cython/C extension for the proven hot function only.
  5. Parallelize last, and match the tool to the shape: threads/ processes on one node, job arrays or MPI on clusters (rseng-hpc-computing), GPUs for data-parallel numeric work (rseng-gpu-computing), distributed frameworks for larger-than- memory data (rseng-big-data-processing).

Preserve correctness at each step: run the tests (rseng-testing) after every optimization, with numerical tolerances where results legitimately differ (bit-identical is often the wrong bar - rseng-numerical-accuracy).

Keep it fast: benchmark tracking

One-off optimization decays; regressions arrive silently in innocent-looking commits. airspeed velocity (asv) runs a benchmark suite across commits, tracks results over time and flags regressions - the performance analogue of a test suite. Start small: benchmark the 3-5 operations users actually wait on, wire the run into CI or a scheduled job (rseng-ci-cd), and treat an unexplained regression like a failing test.

For quick comparisons during work, timeit/hyperfine with repeated runs and reported variance beat single stopwatch numbers; pin the environment (rseng-reproducible-environments) so timings compare across machines honestly - and record the hardware in any published benchmark.

Scaling measurements

Before requesting bigger allocations, measure scaling: run the same problem at 1, 2, 4, 8 workers (strong scaling) or grow the problem with workers (weak scaling), and plot efficiency. Recommending more hardware without a scaling curve is guesswork; the curve also feeds honest resource requests (rseng-hpc-computing) and energy accounting (rseng-green-computing).

Working with this skill

This skill is source-independent: its authority is the profiler and benchmark-tool documentation linked below and standard performance engineering practice.

Learn more (verified):

<!-- related-skills:begin -->

Related skills

Check whether any of these applies before moving on:

  • rseng-big-data-processing - memory-bound escalation path
  • rseng-ci-cd - benchmark regression tracking in CI
  • rseng-gpu-computing - GPU port only after profiling evidence
  • rseng-green-computing - speedups cut energy proportionally
  • rseng-hpc-computing - scaling curves before big allocations
  • rseng-numerical-accuracy - tolerances when optimizations shift results
<!-- related-skills:end -->

리뷰

아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.

댓글

로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.

아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.

보내면 바로 공개됩니다. 이메일 주소나 전화번호 같은 개인 정보는 쓰지 마세요. URL은 2개까지입니다.

0 / 1,000자

이 양식은 reCAPTCHA로 보호되며 Google 개인정보처리방침 및 서비스 약관이 적용됩니다.

같은 저장소의 스킬

rseng-agent-security

무료

Covers operating AI coding agents securely: auditing whether the agent runs sandboxed, permission and dangerous-command configuration, containerized environments, secrets kept out of context, network egress and token privileges, prompt-injection risk, and the privacy of agent use itself. Use PROACTIVELY when permission gating is broadly disabled (skip-permissions modes), when secrets are visible to the agent's shell, or when the agent processes untrusted code, issues or web content; when a project holds personal or confidential data while a cloud agent is active (what leaves the machine, retention, training use), or when the user asks how to run coding agents safely or privately. (Supply-chain security: rseng-security; disclosing agent contributions: rseng-ai-declaration.)

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

fdiblen/rseng-agent-skills242026년 9월 23일 업데이트

rseng-ai-declaration

무료

Covers declaring AI involvement with the AI Declaration Format (https://ai-declaration.org): creating and maintaining an aidecl.yaml that records which AI tools and agents were used, what each contributed, when and how much, plus the README disclosure footnote, JSON/JSON-LD export and CI validation. Machine-readable disclosure is the practice; this format is one implementation. Use PROACTIVELY when an AI assistant or agent creates or modifies a project's content - generate aidecl.yaml if missing, update it after AI-assisted changes. Also use when the user asks about AI transparency, provenance, disclosure or EU AI Act disclosure, wants to state that NO AI was used, or mentions aidecl. For concealment requests see rseng-honesty; for the human review duty see rseng-human-verification.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

fdiblen/rseng-agent-skills242026년 9월 23일 업데이트

rseng-archiving

무료

Covers long-term archiving of research software and data: Software Heritage save requests and SWHID persistent identifiers, Zenodo deposits with versioned DOIs and forge integration, choosing domain and institutional repositories, deciding what to archive (code, data, environments, documentation) and when, and archiving at project retirement. Use when the user wants software or data preserved beyond the life of a forge account, grant or lab, mentions Software Heritage, SWHIDs, Zenodo deposits or archiving, retires or hands over a project, prepares artifacts that must stay resolvable for a paper, or when funder policy requires long-term preservation. (Cutting the release itself is rseng-publishing-releasing; day-to-day data storage and repository choice is rseng-data-management.)

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

fdiblen/rseng-agent-skills242026년 9월 23일 업데이트

rseng-big-data-processing

무료

Covers processing research data that outgrows one machine's memory: out-of-core and chunked computation, Dask for scaling the scientific Python stack, Spark for distributed tabular pipelines, lazy evaluation, partitioning strategies, idempotent and restartable batch jobs, and knowing when NOT to distribute. Use when datasets no longer fit in memory, when the user mentions Dask, Spark, out-of-core or larger-than-memory data, when a pandas/NumPy workflow hits memory limits, or when designing batch pipelines over many files. (Cluster job submission and job arrays are rseng-hpc-computing; pipeline orchestration engines are rseng-workflows; profile first with rseng-performance-profiling.)

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

fdiblen/rseng-agent-skills242026년 9월 23일 업데이트

rseng-ci-cd

무료

Covers continuous integration and delivery for research software: CI/CD concepts, automating builds and tests with GitHub Actions and GitLab CI/CD, and wiring an organization's GitLab CI infrastructure to a GitHub-hosted project. Use when the user asks to set up CI, write a pipeline or workflow file (.github/workflows, .gitlab-ci.yml), add automated builds/tests on push or pull request, publish a package or deploy docs when a tag or release is cut, cache dependencies to speed up CI runs, choose between GitHub Actions and GitLab CI, use self-hosted or GPU runners, mirror a repo, or report external CI status back to GitHub. For what to test and shrinking a test matrix see rseng-testing; for linter and pre-commit policy see rseng-code-quality.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

fdiblen/rseng-agent-skills242026년 9월 23일 업데이트

rseng-citation-hygiene

무료

Covers verifying that every citation is real, correct and current: checking references in manuscripts, READMEs, references files and code metadata against Crossref and OpenAlex, screening cited DOIs against the Retraction Watch database, catching fabricated or mis-attributed citations (a documented AI failure mode), and keeping bibliographies and CITATION.cff files resolvable. Use PROACTIVELY before any bibliography, reference list or citation metadata is finalized or published, whenever the agent itself has produced citations, and when the user asks to check references, mentions broken DOIs, retracted papers or citation verification, or prepares a manuscript, README or software paper with references. (Making your own software citable is rseng-citation-metadata.)

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

fdiblen/rseng-agent-skills242026년 9월 23일 업데이트

fdiblen의 스킬 모두 보기

이 스킬의 문제 신고