pyAlexS3

OpenAlex S3 → DuckDB loader powered by rich progress bars.

Reads OpenAlex NDJSON dumps directly from S3 via DuckDB's httpfs extension — no downloading required.

Features

• 🚀 Direct S3 reads via DuckDB httpfs — no local downloads
• 🦆 Zero-setup DuckDB loading via read_json_auto(...)
• 🎯 Filter by date range (YYYY-MM-DD) and by part numbers
• 🔁 Resume from a specific date and part after a failure
• 🔎 Optional SQL-style WHERE predicate
• 📊 Optional rich progress bar showing batch progress

Installation

pip install pyalexs3

or with uv:

uv add pyalexs3

Python 3.10+ is required.

Quick Start

from pyalexs3.core import OpenAlexS3Processor

p = OpenAlexS3Processor(n_workers=4)

for file_batch, rel in p.lazy_load(
    obj_type="works",
    start_date="2025-01-01",
    end_date="2025-03-01",
    columns=["id", "title", "publication_year"],
):
    df = rel.df()
    print(df.head())

Filter with WHERE clause

for file_batch, rel in p.lazy_load(
    obj_type="works",
    start_date="2025-01-01",
    end_date="2025-03-01",
    columns=["id", "title", "publication_year"],
    where_clause="title IS NOT NULL AND language='en'",
):
    df = rel.df()

Resume After Failure

If your pipeline fails midway, resume from a specific date and part number:

for file_batch, rel in p.lazy_load(
    obj_type="works",
    start_date="2025-01-01",
    end_date="2025-03-01",
    resume_from="2025-01-15/5",  # skip everything before 2025-01-15 part 5
):
    df = rel.df()

Load Specific Parts Only

for file_batch, rel in p.lazy_load(
    obj_type="works",
    start_date="2025-01-01",
    end_date="2025-01-01",
    parts=[0, 1, 2],  # only load part_000.gz, part_001.gz, part_002.gz
):
    df = rel.df()

Show Progress

p = OpenAlexS3Processor(n_workers=4, show_progress=True)

for file_batch, rel in p.lazy_load(obj_type="works"):
    df = rel.df()

Track Which Files Were Processed

Each lazy_load iteration yields both the file batch and the relation:

for file_batch, rel in p.lazy_load(obj_type="works"):
    print(f"Processing: {file_batch}")  # list of S3 keys in this batch
    df = rel.df()

API

OpenAlexS3Processor(n_workers=4, **kwargs)

lazy_load(...) -> Generator[tuple[list[str], DuckDBPyRelation], None, None]

Yields tuple[list[str], duckdb.DuckDBPyRelation]:

• list[str] — S3 keys in this batch (useful for progress tracking)
• DuckDBPyRelation — query the batch with .df(), .arrow(), .fetchall()

Supported Object Types

works, authors, sources, institutions, topics, keywords, publishers, funders, concepts

Behavior & Notes

• No downloads — data is read directly from S3 via DuckDB httpfs. No temp files, no cleanup needed.
• DuckDB — installs and loads httpfs automatically on init. Sets PRAGMA threads to n_workers.
• Object cache — PRAGMA enable_object_cache=true is set by default for repeated queries on the same files.
• S3 auth — OpenAlex S3 is public. No credentials needed.

Testing

Dev dependencies include pytest.

uv sync --extra dev
uv run pytest -q

Tests mock the S3 client directly using unittest.mock to test the file listing and filtering logic without hitting real S3.

Development

• Source layout: src/pyalexs3/
• Typed package marker: src/pyalexs3/py.typed

License

MIT © EurekAI

Citation

If you are using this for research purposes please use this BibTeX for citation:

@misc{pyalexs32025,
    author = {Adityam Ghosh},
    title = {pyalexs3},
    howpublished = {\url{https://github.com/EurekAI-Org/pyalexs3}},
    year = {2025},
    note = {[Accessed 09-10-2025]},
}