Skip to content

QUICKSTART.md

Latest commit

 

History

History
155 lines (115 loc) · 3.61 KB

QUICKSTART.md

File metadata and controls

155 lines (115 loc) · 3.61 KB
Raw

Quick Start Guide - Paper Decomposer

Installation

  1. Install the package with paper decomposer dependencies:
pip install -e ".[paper_decomposer]"

Or install dependencies manually:

pip install nbformat pymupdf papermill pdfminer.six
  1. Ensure Docker is installed and running:
docker --version
docker ps

Quick Test

1. Run Unit Tests

# Test decomposition logic
python -m pytest tests/paper_decomposer/test_decompose.py -v

# Test notebook assembly  
python -m pytest tests/paper_decomposer/test_notebook_assembly.py -v

# Test integration (requires Docker)
python -m pytest tests/paper_decomposer/test_integration_run.py -v

2. Use the Example Script

# Create a simple test script
from paper_decomposer import PaperToNotebookController

# Initialize
controller = PaperToNotebookController(toy_mode=True)

# Test with sample paper (you'll need to create or download a PDF)
result = controller.run_from_pdf("examples/sample_paper.pdf", experiment_index=0)

print(f"Success: {result['success']}")
print(f"Output: {result.get('output_directory')}")

3. Command Line Test

# Download a simple arXiv paper
python -m paper_decomposer.controller \
    https://arxiv.org/abs/1706.03762 \
    --experiment 0 \
    --toy \
    --max-iterations 3

Expected Output

After a successful run, you should see:

output/
  <paper-id>/
    decomposition.json
    notebook-experiment0.ipynb
    notebook-experiment0_executed.ipynb  # if execution succeeded
    trajectory.json
    run_report.json

Troubleshooting

"Import fitz could not be resolved"

pip install pymupdf

"Import nbformat could not be resolved"

pip install nbformat

"Docker execution error"

  • Ensure Docker is running: docker ps
  • Check Docker permissions
  • Try pulling the image manually: docker pull python:3.11-slim

"RLM client not configured"

  • Set up your LLM API keys (Anthropic, OpenAI, etc.)
  • Or configure RLM client explicitly:
from rlm import RLM
from rlm.clients.anthropic import AnthropicClient

rlm = RLM(client=AnthropicClient())
controller = PaperToNotebookController(rlm_client=rlm)

Next Steps

  1. Read the full documentation: paper_decomposer/docs/README.md
  2. Check the feature overview: paper_decomposer/FEATURE_OVERVIEW.md
  3. Explore example code: examples/paper_decomposer_example.py
  4. Try processing your own papers!

Minimal Working Example

#!/usr/bin/env python3
"""Minimal example - just test the pipeline without execution."""

from paper_decomposer.ingest import extract_text_from_pdf, chunk_text
from paper_decomposer.decompose import decompose_paper
from paper_decomposer.notebook_gen import generate_notebook_cells, assemble_notebook
from rlm import RLM

# 1. Extract text
text = extract_text_from_pdf("paper.pdf")
chunks = chunk_text(text, max_chunk_size=8000)

# 2. Decompose
rlm = RLM()
decomposition = decompose_paper(rlm, chunks)
print(f"Found {len(decomposition['experiments'])} experiments")

# 3. Generate notebook
if decomposition['experiments']:
    experiment = decomposition['experiments'][0]
    cells = generate_notebook_cells(rlm, experiment, toy_mode=True)
    assemble_notebook(cells, "output/test_notebook.ipynb")
    print("Notebook created: output/test_notebook.ipynb")

Save this as test_pipeline.py and run:

python test_pipeline.py

Success Criteria

✅ Tests pass (at least the non-Docker ones) ✅ Can decompose a paper to JSON ✅ Can generate a notebook from experiment spec ✅ Can assemble a valid .ipynb file ✅ (Optional) Can execute notebook in Docker