vectorspacelab/arex-skill/annealed-langevin-sampler

Claude Code and agent skill by vectorspacelab/arex-skill

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Run annealed Langevin dynamics with composed scores for F-NPSE or PF-NPSE posterior sampling.

Install

npx vskill@latest install vectorspacelab/arex-skill/annealed-langevin-sampler

SKILL.md

The same SKILL.md was published earlier as vectorspacelab/arex-skill/annealed-langevin-sampler.

Annealed Langevin Sampler

Use this skill when a recovery needs to sample from a posterior approximation using a score function already composed from F-NPSE or PF-NPSE terms. It implements the paper's Algorithm 1 in a bounded, deterministic-seeded form suitable for tests and recovery harnesses.

Do not use this skill to train score networks or to define the F-NPSE algebra. It consumes a callable score and a reference sampler.

Inputs

  • Initial samples from the high-noise reference distribution, or a reference sampler.
  • Composed score callable score(theta, t).
  • Time levels ordered from high noise to low noise.
  • Step size, Langevin steps per level, and random seed.

Outputs

  • Final posterior samples.
  • Trace metadata with score norms, sample means, level count, step size, and seed.

Workflow

  1. Initialize samples from the reference distribution.
  2. Traverse time levels from high noise to low noise.
  3. Apply theta <- theta + 0.5 * step_size * score(theta, t) + sqrt(step_size) * noise.
  4. Log per-level summary statistics.
  5. Return samples and trace.

Validation

Run:

bash
python /share/project/yuyang/workspace/Paper2Skills/Distiller/skills/module-to-skill/scripts/validate_skill_tree.py <skill_dir> --run-tests

The tests use an exact Gaussian score and verify that samples move toward the target mean while preserving trace metadata.

Limitations

The sampler is sensitive to step size and score scaling. Recovery scripts should expose these settings and avoid interpreting a short bounded run as a full-scale reproduction of the paper's 400-level experiments.

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
339
Repository forks
26
7-day trend score
0

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/vectorspacelab/arex-skill/annealed-langevin-sampler/badge)](https://verified-skill.com/skills/vectorspacelab/arex-skill/annealed-langevin-sampler)

Something wrong with this skill or its page? Tell us.