Skip to content

CLAUDE.md

Latest commit

 

History

History
57 lines (44 loc) · 2.96 KB

CLAUDE.md

File metadata and controls

57 lines (44 loc) · 2.96 KB
Raw

Project Guidelines

IMPORTANT: All commits and code must be in English only. Do not write Korean in any commit messages, comments, variable names, or documentation.

Project Root

The root of this project is /Users/comcom/Desktop/papers-with-claudecode/knowledge-graph-builder.

Reference Codebase

  • When expressions like "original intent", "existing", or "existing implementation" appear, explore under /Users/comcom/Desktop/papers-with-claudecode/knowledge-graph-builder/knowledge_graph_builder to understand the original intent.
  • Files in that directory must never be modified. Read (reference) only.

Analyzer Architecture (analyzer/)

A newly developed general-purpose repo analyzer. Extends the HF-specific structure of knowledge_graph_builder/ to analyze arbitrary paper repositories.

UniversalRepoAnalysis Field Definitions

Field Type Description
components list[ComponentInfo] List of classes/functions (name, path, type, metadata)
commits list[CommitInfo] Keyword-filtered commits (sha, date, message, tags)
documentation list[DocumentationInfo] README/docs summaries (path, title, summary, category)
structure dict Model class hierarchy: {ClassName: {inherits: [...], file: path}}
dependencies dict Package dependencies: {frameworks, domain_libs, data, other, raw, source_files}
extensions dict Repo-type-specific data (HF: extensions.models)

structure Design Intent

  • This is not a directory tree
  • Represents Python class inheritance/composition relationships → model architecture hierarchy
  • Distinguished from components (flat list): structure represents inter-class relationships
  • Can be used for automatic placement of prerequisite nodes in learning paths
  • HuggingFaceAnalyzer: scans only modeling_utils.py + top 10 representative models (size limit)
  • GenericAnalyzer: scans all Python files up to 50 (size limit)

dependencies Design Intent

  • Parses requirements.txt, pyproject.toml, setup.py, package.json
  • Category classification:
    • frameworks: torch, jax, tensorflow, etc. → framework prerequisites
    • domain_libs: peft, flash-attn, diffusers, etc. → domain-specific prerequisites
    • data: numpy, datasets, pandas, etc.
    • other: everything else
  • Used for automatic extraction of "what you need to know before reading this paper" in learning paths
  • Implementation location: BaseRepoAnalyzer.scan_dependencies() (shared by all analyzers)

Analyzer File Structure

analyzer/
├── base.py          # BaseRepoAnalyzer: shared scan_dependencies() implementation
├── models.py        # UniversalRepoAnalysis data models
├── registry.py      # @register_analyzer decorator
└── analyzers/
    ├── huggingface.py  # HF Transformers-specific (structure: top 10 representative models)
    └── generic.py      # General-purpose fallback (structure: up to 50 files)