code/tests/test_blocking_smoke.py

"""End-to-end smoke test for src.blocking and src.make_candidate_tsv on a tiny synthetic dataset.

Run with `python tests/test_blocking_smoke.py` (or pytest). Exercises the scipy fallback backend,
the sparse_dot_topn backend when importable, --max-s1 sampling, and the candidate TSV helper.
The data contains the hard cases from docs/dataset_findings.md: an Indic-script vendor name, a
domain-form name, a complete rename (both with matching addresses), an empty vendor address,
templated name collisions, orphans, singletons and a third country.
"""
import json
import os
import random
import subprocess
import sys
import tempfile
from pathlib import Path

ROOT = Path(__file__).resolve().parents[1]
HEADER = "entity_id\tbusiness_name\tbusiness_address\tcountry\n"

US_A = ["Holloway", "Prime", "Westbrook", "Summit", "Riverside", "Oakridge", "Maple", "Cedar", "Liberty", "Pioneer",
        "Harbor", "Evergreen", "Golden", "Granite", "Falcon", "Keystone", "Bluebird", "Ironwood", "Sterling", "Magnolia"]
US_B = ["Dental", "Pediatric", "Seafood", "Consulting", "Logistics", "Realty", "Bakery", "Auto", "Learning", "Fitness"]
US_C = ["Group", "LLC", "Inc", "Services", "Center", "Co"]
US_ST = ["Elm", "Oak", "Westchester", "Montebello", "Ellis", "Fawn", "Center", "Lake", "Hill", "Pine", "Cedar", "Sunset"]
US_TY = [("Street", "St"), ("Road", "Rd"), ("Drive", "Dr"), ("Avenue", "Ave"), ("Court", "Ct")]
US_CITY = [("High Point", "NC"), ("Phoenix", "AZ"), ("Tahlequah", "OK"), ("Fort Worth", "TX"), ("Morganton", "NC"), ("Spicewood", "TX")]
IN_A = ["Shree", "Aditya", "Krishna", "Ganesh", "Lakshmi", "Balaji", "Durga", "Sai", "Om", "Vishnu", "Maruti", "Surya"]
IN_B = ["Marketing", "Properties", "Traders", "Enterprises", "Textiles", "Pharma", "Constructions", "Motors", "Foods"]
IN_C = ["Private Limited", "LLP", "Pvt Ltd", "Limited"]
IN_LOC = ["Gulmohar Colony", "Subhash Nagar", "MG Road", "Civil Lines", "Banjara Hills", "Salt Lake"]
IN_CITY = [("Bhopal", "Madhya Pradesh"), ("New Delhi", "Delhi"), ("Bengaluru", "Karnataka"), ("Hyderabad", "Telangana"),
           ("Kolkata", "West Bengal"), ("Pune", "Maharashtra")]
FR_A = ["Boulangerie", "Atelier", "Maison", "Cabinet", "Garage", "Pharmacie", "Librairie", "Brasserie"]
FR_B = ["Dupont", "Martin", "Lefèvre", "Moreau", "Bernard", "Petit", "Rousseau", "Fontaine"]
FR_C = ["SARL", "SAS", "EURL", "SA"]
FR_ST = ["de la Paix", "Victor Hugo", "des Lilas", "Pasteur", "du Moulin", "Jean Jaurès"]
FR_CITY = [("Lille", "Hauts-de-France"), ("Bordeaux", "Nouvelle-Aquitaine"), ("Nantes", "Pays de la Loire")]


def _typo(s, rng):
    if len(s) < 5:
        return s
    i = rng.randrange(1, len(s) - 2)
    return s[:i] + s[i + 1] + s[i] + s[i + 2:]


def _make_entity(country, rng):
    if country == "US":
        name = f"{rng.choice(US_A)} {rng.choice(US_B)} {rng.choice(US_C)}"
        full, ab = rng.choice(US_TY)
        city, st = rng.choice(US_CITY)
        street = f"{rng.choice(US_ST)} {full}"
        return name, f"{rng.randint(10, 19999)} {street}, {city}, {st}", (full, ab)
    if country == "India":
        name = f"{rng.choice(IN_A)} {rng.choice(IN_B)} {rng.choice(IN_C)}"
        city, st = rng.choice(IN_CITY)
        return name, f"{rng.randint(1, 999)}/{rng.randint(1, 99)}, {rng.choice(IN_LOC)}, {city}, {st}", ("Road", "Rd")
    name = f"{rng.choice(FR_A)} {rng.choice(FR_B)} {rng.choice(FR_C)}"
    city, reg = rng.choice(FR_CITY)
    return name, f"{rng.randint(1, 200)} rue {rng.choice(FR_ST)}, {city}, {reg}", ("rue", "R.")


def _noisy(name, addr, abbr, rng):
    r = rng.random()
    if r < 0.2:
        name = name.upper()
    elif r < 0.35:
        name = _typo(name, rng)
    elif r < 0.45:
        name = "-- " + name
    elif r < 0.55:
        name = " ".join(name.split()[:-1]) or name
    r = rng.random()
    if r < 0.3:
        addr = addr.replace(abbr[0], abbr[1]).upper()
    elif r < 0.4:
        parts = addr.split(", ")
        addr = ", ".join(parts[:1] + ["null"] + parts[1:])
    elif r < 0.5:
        addr = ", ".join(addr.split(", ")[:-1])
    elif r < 0.53:
        addr = ""
    return name, addr


def make_dataset(d: Path, seed: int = 7, n_per_country: int = 90):
    rng = random.Random(seed)
    s1, s2, s3, gt = [], [], [], []
    ids = iter(rng.sample(range(10_000, 99_999), 5000))
    special = {}

    def add_entity(country, name, addr, abbr, copies):
        sid = f"S1-{next(ids)}"
        s1.append((sid, name, addr, country))
        matched = []
        for src, (vn, va) in copies:
            vid = f"{src}-{next(ids)}"
            (s2 if src == "S2" else s3).append((vid, vn, va, country))
            matched.append(vid)
        gt.append((sid, ",".join(matched)))
        return sid, matched

    for country in ("US", "India", "France"):
        for i in range(n_per_country):
            name, addr, abbr = _make_entity(country, rng)
            if i < 8 and country == "US":  # templated name collision block
                name = "Primary Care Group"
            n_copies = rng.choice([0, 1, 2, 3, 3, 4])
            copies = [(rng.choice(["S2", "S3"]), _noisy(name, addr, abbr, rng)) for _ in range(n_copies)]
            add_entity(country, name, addr, abbr, copies)
        for src in ("S2", "S3"):  # orphans
            for _ in range(40):
                name, addr, abbr = _make_entity(country, rng)
                (s2 if src == "S2" else s3).append((f"{src}-{next(ids)}", *_noisy(name, addr, abbr, rng), country))

    special["indic"] = add_entity("India", "Ram Marketing Private Limited", "KH NO. 570/13, West Delhi, New Delhi, Delhi", None,
                                  [("S2", ("राम मार्केटिंग प्राइवेट लिमिटेड", "KH NO. -570/13, WEST DELHI, NEW DELHI, Delhi"))])
    special["domain"] = add_entity("US", "Dutta International School", "1795 Westchester Drive, High Point, NC", None,
                                   [("S3", ("duttainternationalschool.com", "1795 Westchester Dr, High Point, NC"))])
    special["rename"] = add_entity("US", "Universal Finance Group", "17560 Ellis Road, Tahlequah, OK", None,
                                   [("S2", ("Avitavowex", "17560 ELLIS RD, TAHLEQUAH, OK"))])
    special["empty_addr"] = add_entity("US", "Holloway Peak Seafood Inc", "105 Elm Street, Morganton, NC", None,
                                       [("S3", ("Holloway Peak Seafood Inc", ""))])
    rng.shuffle(s2)
    rng.shuffle(s3)
    for fn, rows in (("s1.tsv", s1), ("s2.tsv", s2), ("s3.tsv", s3)):
        with open(d / fn, "w", encoding="utf-8") as f:
            f.write(HEADER)
            f.writelines("\t".join(r) + "\n" for r in rows)
    with open(d / "gt.tsv", "w", encoding="utf-8") as f:
        f.write("source1_entity_id\tmatched_entity_ids\n")
        f.writelines(f"{a}\t{b}\n" for a, b in gt)
    return special


def _simple_norm(x):
    import re
    import unicodedata

    x = "".join(c for c in unicodedata.normalize("NFKD", x) if not unicodedata.combining(c)).lower()
    return re.sub(r"[^\w]+", " ", x).strip()


def write_norm_parquets(d: Path, special) -> Path:
    """Mimic scripts/preprocess_normalize.py output (lowercased country, list-typed address
    tokens, Indic name transliterated with legal words dropped, one null name_translit)."""
    import polars as pl

    nd = d / "norm"
    nd.mkdir(exist_ok=True)
    indic_vid = special["indic"][1][0]
    domain_vid = special["domain"][1][0]
    for tbl, fn in (("s1", "s1.tsv"), ("s2", "s2.tsv"), ("s3", "s3.tsv")):
        rows = [ln.split("\t") for ln in (d / fn).read_text(encoding="utf-8").split("\n")[1:] if ln]
        recs = []
        for eid, name, addr, country in rows:
            core = _simple_norm(name)
            translit = core
            if eid == indic_vid:
                translit = "ram marketing"  # what the lexicon would give
            if eid == domain_vid:
                translit = None  # exercise the name_core fallback
            recs.append({
                "entity_id": eid, "country": country.lower(), "name_core": core, "name_translit": translit,
                "name_spaceless": core.replace(" ", ""),
                "addr_tokens_translit": [t for t in _simple_norm(addr).split() if t != "null"],
                "raw_name": name, "raw_address": addr,
            })
        df = pl.DataFrame(recs, schema={"entity_id": pl.Utf8, "country": pl.Utf8, "name_core": pl.Utf8,
                                        "name_translit": pl.Utf8, "name_spaceless": pl.Utf8,
                                        "addr_tokens_translit": pl.List(pl.Utf8), "raw_name": pl.Utf8,
                                        "raw_address": pl.Utf8})
        df.write_parquet(nd / f"{tbl}.parquet")
        df.write_parquet(nd / f"t{tbl[1]}.parquet")  # same content as a stand-in test split
    return nd


def _run(args, cwd=ROOT):
    env = dict(os.environ, PYTHONPATH=str(ROOT))
    if sys.platform == "darwin":  # fork-based worker pools need this on macOS only
        env.setdefault("OBJC_DISABLE_INITIALIZE_FORK_SAFETY", "YES")
    r = subprocess.run([sys.executable, *args], cwd=cwd, env=env, capture_output=True, text=True)
    if r.returncode != 0:
        print(r.stdout[-5000:], r.stderr[-5000:])
        raise AssertionError(f"command failed: {args}")
    return r.stdout


def _has_sdt():
    try:
        import sparse_dot_topn  # noqa: F401  (availability check)
        return True
    except ImportError:
        return False


def run_case(d: Path, special, backend, extra=(), tag=None, inputs=None):
    import polars as pl

    tag = tag or backend
    out = d / f"cands_{tag}.parquet"
    inputs = inputs or ["--s1", str(d / "s1.tsv"), "--s2", str(d / "s2.tsv"), "--s3", str(d / "s3.tsv")]
    stdout = _run(["-m", "src.blocking", *inputs,
                   "--gt", str(d / "gt.tsv"), "--out", str(out), "--backend", backend, "--workers", "2",
                   "--max-df", "0.1", *extra])
    m = json.loads((d / f"cands_{tag}.metrics.json").read_text())
    ev = m["eval"]
    union = ev["pair_recall"]["overall"]["union"]
    assert isinstance(union, float) and 0.0 <= union <= 1.0
    for f in ("f_name", "f_nameaddr", "f_addr", "f_word", "from_reverse"):
        assert isinstance(ev["pair_recall"]["overall"][f], float)
    assert {c.lower() for c in ev["pair_recall"]["by_country"]} <= {"us", "india", "france"}
    assert set(ev["pair_recall"]["by_source"]) == {"S2", "S3"}
    assert m["cands_per_s1"]["max"] >= 1 and 0 < m["reduction_ratio"] < 1
    assert "k_x1.0" in ev["sweeps"] and "revcap_0" in ev["sweeps"] and "nameaddr" in ev["recall_at_k"]
    if "--reverse-cap" not in extra:  # the full-K sweep config must reproduce the run itself
        assert abs(ev["sweeps"]["k_x1.0"]["union_pair_recall"] - union) < 1e-9
        assert ev["sweeps"]["k_x1.0"]["pairs"] == m["n_pairs"]
    c = pl.read_parquet(out)
    for col in ("rank_name", "rank_nameaddr", "rank_addr", "rank_word", "rank_rev", "is_match"):
        assert col in c.columns, col
    assert (c.get_column("f_name") == c.get_column("rank_name").is_not_null()).all()
    assert c.select(["s1_id", "cand_id"]).is_duplicated().sum() == 0
    assert c.filter(~pl.col("cand_id").str.starts_with("S2-") & ~pl.col("cand_id").str.starts_with("S3-")).height == 0
    got = set(zip(c.get_column("s1_id").to_list(), c.get_column("cand_id").to_list()))
    sampled = "--max-s1" in extra
    queried = set(c.get_column("s1_id").to_list())
    for k, (sid, matched) in special.items():
        if sampled and sid not in queried:
            continue
        for vid in matched:
            assert (sid, vid) in got, f"{tag}: special case {k} pair ({sid},{vid}) missing"
    print(f"[{tag}] union pair recall={union:.4f} entity_any={ev['entity_any_found']['overall']['union']:.4f} "
          f"pairs={m['n_pairs']} cands/S1={m['cands_per_s1']['mean']} rr={m['reduction_ratio']:.4f}")
    return m, out, stdout


def test_smoke():
    import polars as pl

    d = Path(tempfile.mkdtemp(prefix="blocktest_"))
    special = make_dataset(d)
    m_scipy, out_scipy, stdout = run_case(d, special, "scipy")
    print(stdout[stdout.find("===== BLOCKING"):] if "BLOCKING" in stdout else stdout[-3000:])
    assert m_scipy["eval"]["pair_recall"]["overall"]["union"] >= 0.9
    if _has_sdt():
        m_sdt, out_sdt, _ = run_case(d, special, "sdt")
        a = m_scipy["eval"]["pair_recall"]["overall"]["union"]
        b = m_sdt["eval"]["pair_recall"]["overall"]["union"]
        assert abs(a - b) < 0.02, (a, b)
        # identical tf-idf, so the two backends must agree on cosines of shared pairs
        j = pl.read_parquet(out_scipy).join(pl.read_parquet(out_sdt), on=["s1_id", "cand_id"], suffix="_b")
        assert j.height > 0.9 * pl.read_parquet(out_scipy).height
        assert (j.get_column("sim_nameaddr") - j.get_column("sim_nameaddr_b")).abs().max() < 1e-4
    else:
        print("sparse_dot_topn not importable: sdt backend skipped")
    m_s, _, _ = run_case(d, special, "scipy", extra=("--max-s1", "120", "--no-fill-sims", "--reverse-cap", "5"), tag="sample")
    assert m_s["n_query_s1"] == 120 and m_s["n_pool_total"] == m_scipy["n_pool_total"]
    assert m_s["eval"]["n_query_s1"] == 120
    cs = pl.read_parquet(d / "cands_sample.parquet")
    fwd = pl.any_horizontal("f_name", "f_nameaddr", "f_addr", "f_word")
    assert cs.filter(~fwd).group_by("s1_id").len().get_column("len").max() <= 5  # reverse cap honoured
    # pre-normalised parquet input (scripts/preprocess_normalize.py layout)
    nd = write_norm_parquets(d, special)
    backend = "sdt" if _has_sdt() else "scipy"
    m_n, out_n, _ = run_case(d, special, backend, tag="normdir", inputs=["--norm-dir", str(nd), "--split", "train"])
    assert m_n["eval"]["pair_recall"]["overall"]["union"] >= 0.9
    assert set(m_n["countries"]) == {"us", "india", "france"}
    cn = pl.read_parquet(out_n)
    assert set(cn.get_column("country").unique().to_list()) == {"us", "india", "france"}
    ind_sid, ind_vids = special["indic"]  # transliterated Indic name now also matches by name
    assert cn.filter((pl.col("s1_id") == ind_sid) & (pl.col("cand_id") == ind_vids[0])).get_column("sim_name")[0] > 0.3
    out_t = d / "cands_normdir_test.parquet"
    _run(["-m", "src.blocking", "--norm-dir", str(nd), "--split", "test", "--out", str(out_t), "--backend", backend,
          "--workers", "2", "--max-df", "0.1"])
    assert pl.read_parquet(out_t).height == cn.height and "is_match" not in pl.read_parquet(out_t).columns
    print(f"[normdir] train union={m_n['eval']['pair_recall']['overall']['union']:.4f}; test split rows={cn.height}")
    # challenge TSV
    tsv = d / "candidate_pairs.tsv"
    _run(["-m", "src.make_candidate_tsv", "--candidates", str(out_scipy), "--s1", str(d / "s1.tsv"), "--out", str(tsv)])
    lines = tsv.read_text(encoding="utf-8").split("\n")
    assert lines[0] == "source1_entity_id\tcandidate_entity_ids"
    rows = [ln.split("\t") for ln in lines[1:] if ln]
    s1_ids = [ln.split("\t")[0] for ln in (d / "s1.tsv").read_text(encoding="utf-8").split("\n")[1:] if ln]
    assert [r[0] for r in rows] == s1_ids
    for r in rows:
        assert len(r) == 2
        ids = [x for x in r[1].split(",") if x]
        assert len(ids) == len(set(ids)) and all(x[:3] in ("S2-", "S3-") for x in ids)
    print(f"candidate_pairs.tsv OK: {len(rows)} rows; data in {d}")


if __name__ == "__main__":
    test_smoke()