-
Notifications
You must be signed in to change notification settings - Fork 1
README Summary
type: source-summary up: "index_markov_chain_rag_tutorial" source: "README" related: "Markov-Retrieval-Tutorial" tags: [readme, overview, entry-point]
One-page summary of the repository's user-facing README, the entry point for anyone cloning the project.
The README presents the project as a small heterogeneous-graph retrieval prototype and points to docs/tutorial.pdf as the primary artifact. The Python code reproduces every number and figure cited in the tutorial.
| Item | Value |
|---|---|
| Embedding model |
all-MiniLM-L6-v2 (used for both chunk text and concept strings) |
| Retrieval method | Parametric Markov chain, Random Walk With Restart |
| Node types | chunks, documents, concepts |
| Corpus size | 12 entities, 24 chunks |
| Setup tool |
uv venv plus uv pip install -r requirements.txt
|
| Entry point | python run_tutorial.py |
| Model cache |
~/.cache/huggingface, about 90MB on first run |
For each query the script prints:
- Vanilla baseline ranking from plain cosine query-chunk similarity
- Starting concepts and chunks selected for the random walk
- Chain ranking of top chunks, documents, concepts by stationary mass
- For tail-test queries, a path-length decomposition for selected entities
The script writes three PDFs to figures/. See Tutorial Figures for details on each.
An earlier README described a code/ subdirectory and a writeup/ directory for the tutorial PDF and figures. Both were retired before this wiki existed; modules now live at the repository root and the tutorial PDF lives at docs/tutorial.pdf with committed figures at docs/figures/. The README has been updated to match.