daily_papers_sync.py

#!/usr/bin/env python3
# /// script
# requires-python = ">=3.10"
# dependencies = [
#     "datasets",
#     "requests",
#     "huggingface-hub",
# ]
# ///
"""
Daily sync for huggingface/trending-papers-x dataset.
Indexes new papers and updates GitHub/project URLs via HF Papers API.

Run locally:  uv run daily_papers_sync.py
Run as HF Job: hf jobs uv run daily_papers_sync.py --secrets HF_TOKEN
"""

from __future__ import annotations

import os
import re
from datetime import datetime, timedelta, timezone
from typing import Any, Optional
from urllib.parse import urlparse

import requests
from datasets import load_dataset

REPO_ID = "huggingface/trending-papers-x"
API_BASE = "https://huggingface.co/api"
HARD_LIMIT = 150
HOURS_LOOKBACK = 24

# Regex patterns for arXiv ID validation
_ARXIV_URL_RE = re.compile(r"https?://(?:www\.)?arxiv\.org/(?:abs|pdf)/(?P<id>[^?#]+)", re.I)
_ARXIV_NEW_RE = re.compile(r"^\d{4}\.\d{4,5}$")


def normalize_arxiv_id(value: Any) -> Optional[str]:
    """Extract and validate arXiv ID from various formats."""
    if not value:
        return None
    s = str(value).strip()

    # Extract from URL if present
    if m := _ARXIV_URL_RE.search(s):
        s = m.group("id")

    s = s.strip().rstrip("/")
    if s.lower().endswith(".pdf"):
        s = s[:-4]
    if s.lower().startswith("arxiv:"):
        s = s[6:]

    # Remove version suffix
    s = re.sub(r"v\d+$", "", s)

    # Validate new-style arXiv ID format
    if not _ARXIV_NEW_RE.fullmatch(s):
        return None

    # Validate month (positions 2-3)
    month = int(s[2:4])
    return s if 1 <= month <= 12 else None


def normalize_github_repo(value: Any) -> Optional[str]:
    """Extract and normalize GitHub repo URL."""
    if not value:
        return None
    s = str(value).strip()

    if s.startswith("[email protected]:"):
        s = f"https://github.com/{s[15:]}"
    elif s.startswith("github.com/"):
        s = f"https://{s}"

    p = urlparse(s)
    if p.scheme not in ("http", "https"):
        return None

    host = (p.netloc or "").lower().removeprefix("www.")
    if host != "github.com":
        return None

    parts = [x for x in p.path.split("/") if x]
    if len(parts) < 2:
        return None

    owner, repo = parts[0], parts[1].removesuffix(".git")
    return f"https://github.com/{owner}/{repo}"


def normalize_url(value: Any) -> Optional[str]:
    """Validate and normalize a URL."""
    if not value:
        return None
    s = str(value).strip()
    p = urlparse(s)
    return s if p.scheme in ("http", "https") and p.netloc else None


def parse_date(value: Any) -> Optional[datetime]:
    """Parse date string into datetime."""
    if isinstance(value, datetime):
        return value.replace(tzinfo=timezone.utc) if value.tzinfo is None else value
    if not value:
        return None

    for fmt in ["%Y-%m-%dT%H:%M:%S.%fZ", "%Y-%m-%dT%H:%M:%SZ", "%Y-%m-%d"]:
        try:
            dt = datetime.strptime(str(value).strip(), fmt)
            return dt.replace(tzinfo=timezone.utc)
        except ValueError:
            continue
    return None


def get_token() -> str:
    """Get HF token from environment or huggingface-cli."""
    if token := os.environ.get("HF_TOKEN", "").strip():
        return token
    try:
        from huggingface_hub import HfFolder
        return (HfFolder.get_token() or "").strip()
    except Exception:
        return ""


def get_paper(session: requests.Session, arxiv_id: str) -> Optional[dict]:
    """Fetch paper from API, returns None if not found."""
    try:
        r = session.get(f"{API_BASE}/papers/{arxiv_id}", timeout=30)
        return r.json() if r.status_code == 200 else None
    except Exception:
        return None


def index_paper(session: requests.Session, arxiv_id: str) -> bool:
    """Index a paper by arXiv ID. Returns True on success."""
    try:
        r = session.post(f"{API_BASE}/papers/index", json={"arxivId": arxiv_id}, timeout=30)
        return r.status_code == 200
    except Exception:
        return False


def update_paper_links(
    session: requests.Session,
    arxiv_id: str,
    github_repo: Optional[str] = None,
    project_page: Optional[str] = None,
) -> bool:
    """Update GitHub repo and/or project page for a paper."""
    payload = {}
    if github_repo:
        payload["githubRepo"] = github_repo
    if project_page:
        payload["projectPage"] = project_page

    if not payload:
        return False

    try:
        r = session.post(f"{API_BASE}/papers/{arxiv_id}/links", json=payload, timeout=30)
        return r.status_code == 200
    except Exception:
        return False


def main() -> None:
    token = get_token()
    if not token:
        print("ERROR: HF token not found. Set HF_TOKEN or run `huggingface-cli login`.")
        exit(1)

    session = requests.Session()
    session.headers.update({
        "Content-Type": "application/json",
        "Authorization": f"Bearer {token}",
    })

    cutoff_time = datetime.now(timezone.utc) - timedelta(hours=HOURS_LOOKBACK)

    print(f"Dataset: {REPO_ID}")
    print(f"Lookback: {HOURS_LOOKBACK}h (since {cutoff_time.strftime('%Y-%m-%d %H:%M UTC')})")
    print(f"Limit: {HARD_LIMIT} papers")
    print("-" * 50)

    dataset = load_dataset(REPO_ID, split="train", streaming=True)

    stats = {"indexed": 0, "github": 0, "project": 0, "not_found": 0, "skipped": 0}
    processed = 0

    for row in dataset:
        if processed >= HARD_LIMIT:
            break

        arxiv_id = normalize_arxiv_id(row.get("arxiv_id") or row.get("paper_id"))
        if not arxiv_id:
            continue

        # Check date
        date_str = row.get("date") or row.get("published_at") or row.get("created_at")
        if (paper_date := parse_date(date_str)) and paper_date < cutoff_time:
            stats["skipped"] += 1
            continue

        # Get links from dataset
        github_repo = normalize_github_repo(row.get("github") or row.get("github_url"))
        project_page = normalize_url(row.get("project_page_url") or row.get("project_page"))

        # Skip if no links
        if not github_repo and not project_page:
            continue

        processed += 1

        # Check if paper exists
        paper = get_paper(session, arxiv_id)
        just_indexed = False

        # Index if not found
        if paper is None:
            if index_paper(session, arxiv_id):
                stats["indexed"] += 1
                just_indexed = True
                print(f"INDEXED: {arxiv_id}")
            else:
                stats["not_found"] += 1
                print(f"SKIP: {arxiv_id} - could not index")
                continue

        # Determine what to update (if just indexed, paper has no links yet)
        has_github = False if just_indexed else bool(paper.get("githubRepo"))
        has_project = False if just_indexed else bool(paper.get("projectPage"))

        github_to_set = github_repo if github_repo and not has_github else None
        project_to_set = project_page if project_page and not has_project else None

        if not github_to_set and not project_to_set:
            print(f"SKIP: {arxiv_id} - already has links")
            continue

        # Update links
        if update_paper_links(session, arxiv_id, github_to_set, project_to_set):
            if github_to_set:
                stats["github"] += 1
                print(f"SET GITHUB: {arxiv_id} -> {github_to_set}")
            if project_to_set:
                stats["project"] += 1
                print(f"SET PROJECT: {arxiv_id} -> {project_to_set}")
        else:
            print(f"ERROR: {arxiv_id} - failed to update links")

    print("-" * 50)
    print(f"Processed: {processed}")
    print(f"Indexed: {stats['indexed']}")
    print(f"GitHub added: {stats['github']}")
    print(f"Project added: {stats['project']}")
    print(f"Not found: {stats['not_found']}")
    print(f"Skipped (old): {stats['skipped']}")


if __name__ == "__main__":
    main()