"""Download, extract, and join PatentsView text files — no local storage. Runs on any machine with network access. Downloads PatentsView text TSVs, extracts with 7-zip (deflate64), joins with IMPACT, pushes enriched parquet to HF Hub. Nothing stays on the local machine. Requirements: pip install huggingface_hub requests pandas tqdm brew install p7zip (macOS) or apt install p7zip-full (Linux) Usage: export HF_TOKEN=hf_... python scripts/cloud/patentsview_pipeline.py --year 2022 """ import argparse import os import subprocess import tempfile from pathlib import Path import pandas as pd import requests from huggingface_hub import HfApi, hf_hub_download from tqdm import tqdm PATENTSVIEW_URLS = { "drawing_desc": "https://s3.amazonaws.com/data.patentsview.org/draw-description-text/g_draw_desc_text_{year}.tsv.zip", "detail_desc": "https://s3.amazonaws.com/data.patentsview.org/detail-description-text/g_detail_desc_text_{year}.tsv.zip", "brief_summary": "https://s3.amazonaws.com/data.patentsview.org/brief-summary-text/g_brf_sum_text_{year}.tsv.zip", "claims": "https://s3.amazonaws.com/data.patentsview.org/claims/g_claims_{year}.tsv.zip", "patent_meta": "https://s3.amazonaws.com/data.patentsview.org/download/g_patent.tsv.zip", } CHUNK = 8 * 1024 * 1024 # 8MB download chunks def download_file(url: str, dest: Path) -> Path: if dest.exists(): print(f" Cached: {dest.name}") return dest print(f" Downloading {dest.name}...") r = requests.get(url, stream=True, timeout=60) r.raise_for_status() total = int(r.headers.get("content-length", 0)) with open(dest, "wb") as f: with tqdm(total=total, unit="B", unit_scale=True, desc=f" {dest.name}") as pbar: for chunk in r.iter_content(chunk_size=CHUNK): f.write(chunk) pbar.update(len(chunk)) return dest def extract_deflate64(zip_path: Path, out_dir: Path) -> Path | None: """Extract using 7-zip (handles deflate64 that Python's zipfile can't).""" result = subprocess.run( ["7z", "x", str(zip_path), f"-o{out_dir}", "-y"], capture_output=True, text=True, ) if result.returncode != 0: print(f" 7z failed: {result.stderr[:200]}") return None # Find the extracted TSV tsv_files = list(out_dir.glob("*.tsv")) return tsv_files[0] if tsv_files else None def main(): parser = argparse.ArgumentParser() parser.add_argument("--year", default="2022") parser.add_argument("--out-repo", default="midah/patent-wireframes") parser.add_argument("--out-file", default="enriched_{year}_full.parquet") args = parser.parse_args() token = os.environ.get("HF_TOKEN") if not token: raise RuntimeError("Set HF_TOKEN") year = args.year out_file = args.out_file.format(year=year) with tempfile.TemporaryDirectory() as tmpdir: tmp = Path(tmpdir) print(f"Working directory: {tmp}") # ── Download and extract PatentsView tables ─────────────────────────── tables = {} for table_name, url_template in PATENTSVIEW_URLS.items(): url = url_template.format(year=year) zip_dest = tmp / url.split("/")[-1] zip_path = download_file(url, zip_dest) print(f" Extracting {zip_path.name}...") tsv_path = extract_deflate64(zip_path, tmp / table_name) if tsv_path: tables[table_name] = pd.read_csv(tsv_path, sep="\t", dtype=str, low_memory=False) print(f" {table_name}: {len(tables[table_name]):,} rows") zip_path.unlink() # free space immediately else: print(f" WARNING: could not extract {table_name}") tables[table_name] = pd.DataFrame() # ── Download existing enriched parquet from Hub ─────────────────────── print("\nDownloading existing enriched parquet from HF Hub...") try: base_parquet = hf_hub_download( repo_id=args.out_repo, filename=f"enriched_{year}.parquet", repo_type="dataset", token=token, ) df = pd.read_parquet(base_parquet) print(f" Loaded {len(df):,} rows") except Exception: print(" No existing parquet — starting from IMPACT CSV") impact_csv = hf_hub_download( repo_id="AI4Patents/IMPACT", filename=f"{year}.csv", repo_type="dataset", token=token, ) df = pd.read_csv(impact_csv) print(f" Loaded {len(df):,} rows from IMPACT CSV") # ── Join text tables ────────────────────────────────────────────────── def agg_text(tdf: pd.DataFrame, id_col: str, text_col: str) -> pd.DataFrame: if tdf.empty or text_col not in tdf.columns: return pd.DataFrame(columns=[id_col, text_col]) return ( tdf.groupby(id_col)[text_col] .apply(lambda x: "\n".join(x.dropna().astype(str))) .reset_index() ) id_col = "patent_id" if "patent_id" in df.columns else "id" df[id_col] = df[id_col].astype(str) for canonical, (tname, tcol) in { "detailed_description": ("detail_desc", "detail_desc_text"), "brief_summary": ("brief_summary", "brf_sum_text"), "claims": ("claims", "claims_text"), }.items(): tdf = tables.get(tname, pd.DataFrame()) if not tdf.empty and tcol in tdf.columns: tdf[id_col] = tdf[id_col].astype(str) agg = agg_text(tdf, id_col, tcol).rename(columns={tcol: canonical}) df = df.merge(agg, on=id_col, how="left") df[canonical] = df[canonical].fillna("") print(f" Joined {canonical}: {(df[canonical] != '').sum():,} non-empty") # Patent metadata (date, type) meta = tables.get("patent_meta", pd.DataFrame()) if not meta.empty: meta[id_col] = meta[id_col].astype(str) meta_cols = [id_col] + [c for c in ["patent_date","patent_type","wipo_kind"] if c in meta.columns] df = df.merge(meta[meta_cols].drop_duplicates(id_col), on=id_col, how="left") print(f" Joined patent_meta: {df['patent_date'].notna().sum():,} dates") # ── Push to Hub ─────────────────────────────────────────────────────── out_path = tmp / out_file df.to_parquet(out_path, index=False) size_mb = out_path.stat().st_size / 1e6 print(f"\nSaving {size_mb:.1f}MB parquet ({len(df):,} rows)...") api = HfApi(token=token) api.upload_file( path_or_fileobj=str(out_path), path_in_repo=out_file, repo_id=args.out_repo, repo_type="dataset", ) print(f"Pushed → hf://datasets/{args.out_repo}/{out_file}") if __name__ == "__main__": main()