| |
| import gzip, json, os, shutil |
| import pyarrow as pa |
| import pyarrow.parquet as pq |
|
|
| ROOT = os.environ.get("UNSC_CORPUS_ROOT", ".") |
| D = f"{ROOT}/publication/data" |
| src = f"{D}/fulltext.jsonl" |
| schema = pa.schema([("record_id", pa.int64()), ("symbol", pa.string()), |
| ("family", pa.string()), ("date", pa.string()), |
| ("n_chars", pa.int64()), ("text", pa.string())]) |
|
|
| writer = pq.ParquetWriter(f"{D}/fulltext.parquet", schema, compression="zstd") |
| batch, N = [], 0 |
| def flush(): |
| global batch |
| if not batch: |
| return |
| cols = {f.name: [r.get(f.name) for r in batch] for f in schema} |
| cols["record_id"] = [int(x) if x is not None else None for x in cols["record_id"]] |
| writer.write_table(pa.table(cols, schema=schema)) |
| batch = [] |
| for line in open(src): |
| batch.append(json.loads(line)); N += 1 |
| if len(batch) >= 2000: |
| flush() |
| flush(); writer.close() |
| print(f"wrote fulltext.parquet ({N} rows)", flush=True) |
|
|
| with open(src, "rb") as f_in, gzip.open(f"{src}.gz", "wb", compresslevel=6) as f_out: |
| shutil.copyfileobj(f_in, f_out, length=1024*1024) |
| os.remove(src) |
| print("wrote fulltext.jsonl.gz; removed uncompressed", flush=True) |
| print("DONE", flush=True) |
|
|
