pgz-sport/scripts/godisnjak_extract.py_prije_env_deepseek

#!/usr/bin/env python3
import os
# ═══════════════════════════════════════════════════════════════════
# Fajl: godisnjak_extract.py
# Verzija: 1.0.0
# Datum: 03.05.2026
# Autor: Damir Radulić <dradulic@outlook.com>
# Lokacija: /opt/pgz-sport/scripts/godisnjak_extract.py
# Svrha: LLM ekstrakcija osoba/uloga iz godisnjaka PGZ (Phase 2)
# Zavisi od: httpx, psycopg2, rapidfuzz
# Utječe na: pgz_sport.clanovi
# ═══════════════════════════════════════════════════════════════════
import asyncio, glob, json, logging, re, sys, time
import httpx, psycopg2
from psycopg2.extras import execute_batch
from rapidfuzz import fuzz

logging.basicConfig(
    level=logging.INFO,
    format="[%(asctime)s] %(message)s",
    datefmt="%H:%M:%S",
    handlers=[
        logging.FileHandler("/opt/pgz-sport/logs/godisnjak_extract.log"),
        logging.StreamHandler(),
    ],
)
log = logging.getLogger("extract")

DSN = f"host=10.10.0.2 port=6432 dbname=rinet_v3 user=rinet password={os.environ['DB_PASSWORD']}"
VLLM_URL = "http://localhost:8001/v1/chat/completions"
VLLM_MODEL = "Qwen/Qwen2.5-7B-Instruct-AWQ"
DATA_DIR = "/opt/pgz-sport/_data/godisnjaci"
MAX_WORKERS = 4
CHUNK_SIZE = 1400

EXTRACT_PROMPT = """Ekstrahiraj iz teksta SVA imena osoba i njihove uloge.
Vrati ISKLJUCIVO valid JSON (bez markdown, bez objasnjenja):
{"osobe": [{"ime":"X","prezime":"Y","klub":"Z","uloga":"igrac","godina_rodenja":1990}]}

Dozvoljene uloge: predsjednik, dopredsjednik, tajnik, blagajnik, clan_uprave,
igrac, sportas, glavni_trener, trener, pomocni_trener, kondicioni_trener,
selektor, izbornik, team_manager, voditelj, lijecnik, fizioterapeut,
kineziolog, maser, sudac, volonter

Pravila:
1. Samo HRVATSKA osobe s punim imenom i prezimenom
2. Ako klub nije eksplicitno naveden -> klub=""
3. NE izmisljaj - samo jasno navedena imena u tekstu
4. Godina rodenja samo ako eksplicitno u tekstu, inace izostavi"""


def chunk_text(text, size=CHUNK_SIZE):
    paragraphs = re.split(r'\n\n+', text)
    chunks, cur = [], ""
    for p in paragraphs:
        if len(cur) + len(p) > size:
            if cur: chunks.append(cur.strip())
            cur = p
        else:
            cur += "\n\n" + p
    if cur: chunks.append(cur.strip())
    return [c for c in chunks if len(c) > 80]


# Preload klub cache
def load_klub_cache(conn):
    cur = conn.cursor()
    cur.execute("SELECT id, naziv FROM pgz_sport.klubovi WHERE aktivan=true OR aktivan IS NULL LIMIT 2000")
    return cur.fetchall()


def fuzzy_klub(naziv, cache):
    if not naziv or len(naziv) < 3:
        return None
    best_id, best_score = None, 0
    for kid, kname in cache:
        score = fuzz.token_set_ratio(naziv.lower(), kname.lower())
        if score > best_score:
            best_score, best_id = score, kid
    return best_id if best_score > 72 else None


async def extract_persons(chunk_text_str, semaphore):
    async with semaphore:
        try:
            async with httpx.AsyncClient(timeout=90.0) as c:
                r = await c.post(VLLM_URL, json={
                    "model": VLLM_MODEL,
                    "messages": [
                        {"role": "system", "content": EXTRACT_PROMPT},
                        {"role": "user", "content": chunk_text_str[:5000]},
                    ],
                    "temperature": 0.05,
                    "max_tokens": 2500,
                    "response_format": {"type": "json_object"},
                })
                d = r.json()
                content = d["choices"][0]["message"]["content"]
                return json.loads(content)
        except Exception as e:
            log.debug(f"Extract fail: {e}")
            return {"osobe": []}


VALID_ULOGE = {
    "predsjednik","dopredsjednik","tajnik","blagajnik","clan_uprave",
    "igrac","sportas","glavni_trener","trener","pomocni_trener","kondicioni_trener",
    "selektor","izbornik","team_manager","voditelj","lijecnik","fizioterapeut",
    "kineziolog","maser","sudac","volonter"
}


async def main():
    conn = psycopg2.connect(DSN)
    conn.autocommit = True
    cur = conn.cursor()

    # Backup
    cur.execute("""CREATE TABLE IF NOT EXISTS pgz_sport.clanovi_pre_godisnjak_backup
        AS SELECT * FROM pgz_sport.clanovi WHERE 1=0""")
    cur.execute("""INSERT INTO pgz_sport.clanovi_pre_godisnjak_backup
        SELECT * FROM pgz_sport.clanovi""")
    log.info("Backup created")

    klub_cache = load_klub_cache(conn)
    log.info(f"Klub cache: {len(klub_cache)} klubova")

    files = sorted(glob.glob(f"{DATA_DIR}/godisnjak_*_layout.txt"))
    log.info(f"Files: {len(files)}")

    semaphore = asyncio.Semaphore(MAX_WORKERS)
    total_inserted = 0
    total_skipped = 0

    for f in files:
        m = re.search(r'godisnjak_(\d{4})_layout', f)
        year = m.group(1) if m else "?"
        with open(f) as fp:
            text = fp.read()

        chunks = chunk_text(text)
        log.info(f"Year {year}: {len(chunks)} chunks")

        tasks = [extract_persons(c, semaphore) for c in chunks]
        results = await asyncio.gather(*tasks)

        year_ins = 0
        rows = []
        for res in results:
            for o in res.get("osobe", []):
                ime = (o.get("ime") or "").strip()
                prezime = (o.get("prezime") or "").strip()
                if not ime or not prezime or len(ime) < 2 or len(prezime) < 2:
                    continue
                # Basic sanity — no numbers, no too-long names
                if re.search(r'\d', ime+prezime) or len(ime+prezime) > 60:
                    continue

                uloga = (o.get("uloga") or "igrac").lower().strip()
                if uloga not in VALID_ULOGE:
                    uloga = "igrac"

                klub_naziv = (o.get("klub") or "").strip()
                klub_id = fuzzy_klub(klub_naziv, klub_cache)

                rows.append((
                    ime, prezime, uloga, klub_id,
                    "godisnjak",
                    json.dumps({"year": int(year), "klub_naziv": klub_naziv}),
                    "sportas",
                ))

        # Batch upsert — ON CONFLICT skip duplicates by ime+prezime+savez_izvor+year via metadata
        for row in rows:
            try:
                cur.execute("""
                    INSERT INTO pgz_sport.clanovi
                        (ime, prezime, uloga, klub_id, savez_izvor, metadata, kategorija)
                    VALUES (%s,%s,%s,%s,%s,%s,%s)
                    ON CONFLICT DO NOTHING
                """, row)
                if cur.rowcount:
                    year_ins += 1
            except Exception as e:
                log.debug(f"Insert skip: {e}")

        total_inserted += year_ins
        log.info(f"  {year}: {year_ins} osoba inserted (running total: {total_inserted})")

    cur.execute("SELECT count(*) FROM pgz_sport.clanovi WHERE savez_izvor='godisnjak'")
    final = cur.fetchone()[0]
    conn.close()

    log.info(f"""
=== EXTRACT DONE ===
Inserted this run: {total_inserted}
Total godisnjak u DB: {final}
""")

    import requests as rq
    rq.post(
        "https://api.telegram.org/bot8535797835:AAFItT-92jzZ9NWFafLxn0dLa1_n2s-JE5Y/sendMessage",
        data={"chat_id": "7969491558",
              "text": f"✅ Godisnjak LLM extract DONE: {total_inserted} novih osoba, {final} total"},
        timeout=10,
    )


if __name__ == "__main__":
    asyncio.run(main())