o2o-negosium-original/negodata/backend/services/statistics_service.py

import uuid

from fastapi import Depends

from common.database.db_session_manager import DB_SESSION_MNG
from common.database.model.models import quotations
from common.enums import DBWRType, ErrorType, QuotationType, SessionStatus, CardType, CloseReason, ChatSender
from common.utils.gtime import GTime
from crud.statistics_crud import StatisticsCRUD, IStatisticsCRUD
from router.v1.statistics.protocol import (
    Res_StatisticsSummary,
    StatScope,
    StatKpi,
    StatMonthPoint,
    StatMarkupPoint,
    StatOutcome,
    StatParticipation,
    StatTypeRow,
    StatCategory,
    StatCardUsage,
)

WINDOW_MONTHS = 6  # 최근 6개월(당월 포함) 창


class StatisticsService:
    """통계(성과 분석) 집계. 회사 전체(company)와 내 견적(mine) 두 스코프를 한 응답으로 내린다.

    전부 파생(저장 안 함) — 조회 때 sessions/items/chats 조인 집계한다. 읽기 전용.
    절감 원천은 '낙찰 세션'(preferred_sp_id) 이며, 여기서 총절감·추이·유형·카테고리·앵커도달률을 모두 파생한다.
    """

    def __init__(self, stat_crud: IStatisticsCRUD = Depends(StatisticsCRUD)):
        self.stat_crud = stat_crud

    async def get_summary(self, company_id: str, user_id: str) -> Res_StatisticsSummary:
        res = Res_StatisticsSummary()
        company_uuid = uuid.UUID(company_id)
        user_uuid = uuid.UUID(user_id)
        labels, window_start = self._window(GTime.UTC())

        res.company = await self._scope(company_uuid, None, labels, window_start)
        res.mine = await self._scope(company_uuid, user_uuid, labels, window_start)
        return res

    # ── 스코프 집계 ─────────────────────────────────────────────
    async def _scope(self, company_uuid, owner_uuid, labels, since) -> StatScope:
        scope = StatScope()

        win_rows = await self._read(lambda s: self.stat_crud.winning_sessions(s, company_uuid, owner_uuid, since))
        outcome_rows = await self._read(lambda s: self.stat_crud.outcome_counts(s, company_uuid, owner_uuid, since))
        type_rows = await self._read(lambda s: self.stat_crud.type_counts(s, company_uuid, owner_uuid, since))
        part_rows = await self._read(lambda s: self.stat_crud.participation_counts(s, company_uuid, owner_uuid, since))
        regen = await self._read_scalar(lambda s: self.stat_crud.regen_avg_round(s, company_uuid, owner_uuid, since))
        markup = await self._read_scalar(lambda s: self.stat_crud.markup_suppression(s, company_uuid, owner_uuid, since))
        markup_rows = await self._read(lambda s: self.stat_crud.markup_suppression_monthly(s, company_uuid, owner_uuid, since))
        card_rows = await self._read(lambda s: self.stat_crud.card_usage(s, company_uuid, owner_uuid, since))
        card_effect_rows = await self._read(lambda s: self.stat_crud.card_effect_chats(s, company_uuid, owner_uuid, since))

        scope.trend = self._trend(win_rows, labels)
        scope.markup_trend = [StatMarkupPoint(month=r[0], rate=float(r[1] or 0.0)) for r in markup_rows]
        scope.categories = self._categories(win_rows)
        scope.outcome = self._outcome(outcome_rows)
        scope.participation = self._participation(part_rows)
        scope.type_split = self._type_split(type_rows, win_rows)
        scope.cards = self._cards(card_rows, self._card_drops(card_effect_rows))
        scope.kpi = self._kpi(win_rows, scope.trend, scope.outcome, regen, markup)
        return scope

    # ── 파생 계산 ───────────────────────────────────────────────
    def _kpi(self, win_rows, trend, outcome, regen, markup) -> StatKpi:
        k = StatKpi()
        total_saving = sum(int(r.target_price) - int(r.bid_price) for r in win_rows)
        total_target = sum(int(r.target_price) for r in win_rows)
        k.total_savings = total_saving
        k.savings_rate = (total_saving / total_target) if total_target else 0.0
        k.anchor_reach_rate = self._anchor_reach(win_rows)

        closed = outcome.awarded + outcome.open_price + outcome.open_equal + outcome.open_noshow + outcome.open_reject
        k.closed_count = closed
        k.award_rate = (outcome.awarded / closed) if closed else 0.0
        k.regen_avg_round = round(regen, 2)
        k.offline_award_count = sum(1 for r in win_rows if r.is_offline)
        k.markup_suppression_rate = round(markup, 4)  # 인상억제율(재협상 직전 라운드 투찰가 대비, 파생)
        # 전월 대비: 마지막 두 달 절감액 차(창에 2개월 미만이면 0).
        k.savings_delta_mom = (trend[-1].savings - trend[-2].savings) if len(trend) >= 2 else 0
        return k

    def _anchor_reach(self, win_rows) -> float:
        # (목표−투찰)/(목표−앵커), 앵커 있고 목표>앵커인 세션만 평균.
        # 세션별로 [0,100%] 클램프 후 평균 — '도달률'이라 앵커 도달=100% 상한(앵커 뚫어도 100%로).
        # (앵커 gap 이 작으면 원비율이 100% 훌쩍 넘어 평균이 왜곡되므로 캡한다.)
        vals = []
        for r in win_rows:
            if r.anchoring_price is None:
                continue
            target, bid, anchor = int(r.target_price), int(r.bid_price), int(r.anchoring_price)
            span = target - anchor
            if span > 0:
                reach = (target - bid) / span
                vals.append(min(1.0, max(0.0, reach)))
        return (sum(vals) / len(vals)) if vals else 0.0

    def _trend(self, win_rows, labels) -> list:
        bucket = {m: {"savings": 0, "target": 0} for m in labels}
        for r in win_rows:
            m = f"{r.updated_at.year:04d}-{r.updated_at.month:02d}"
            if m in bucket:
                bucket[m]["savings"] += int(r.target_price) - int(r.bid_price)
                bucket[m]["target"] += int(r.target_price)
        out = []
        for m in labels:
            b = bucket[m]
            rate = (b["savings"] / b["target"]) if b["target"] else 0.0
            out.append(StatMonthPoint(month=m, savings=b["savings"], rate=rate))
        return out

    def _categories(self, win_rows) -> list:
        # TODO: items.category 자유텍스트 그룹 — 표기 흔들리면 지저분. 카테고리 정규화(코드/테이블) 후 개선.
        agg: dict = {}
        for r in win_rows:
            key = r.category or "미분류"
            a = agg.setdefault(key, {"savings": 0, "count": 0})
            a["savings"] += int(r.target_price) - int(r.bid_price)
            a["count"] += 1
        rows = [StatCategory(category=k, savings=v["savings"], count=v["count"]) for k, v in agg.items()]
        rows.sort(key=lambda x: x.savings, reverse=True)
        return rows

    def _outcome(self, outcome_rows) -> StatOutcome:
        by = {int(cr): int(n) for cr, n in outcome_rows if cr is not None}
        return StatOutcome(
            awarded=by.get(CloseReason.AWARDED.value, 0),
            open_price=by.get(CloseReason.OPEN_PRICE.value, 0),
            open_equal=by.get(CloseReason.OPEN_EQUAL.value, 0),
            open_noshow=by.get(CloseReason.OPEN_NOSHOW.value, 0),
            open_reject=by.get(CloseReason.OPEN_REJECT.value, 0),
        )

    def _participation(self, part_rows) -> StatParticipation:
        by = {int(st): int(n) for st, n in part_rows if st is not None}
        return StatParticipation(
            bid=by.get(SessionStatus.DONE.value, 0),
            no_participate=by.get(SessionStatus.NOT_PARTICIPATED.value, 0),
            rejected=by.get(SessionStatus.REJECTED.value, 0),
        )

    def _type_split(self, type_rows, win_rows) -> list:
        # 4개 코드(협상 1·3 / 견적 2·4=1:N)를 2그룹으로 묶는다. 낙찰률·건수=type_counts, 평균절감=낙찰세션.
        grp = {"nego": {"count": 0, "awarded": 0}, "auction": {"count": 0, "awarded": 0}}
        for t, cnt, awarded in type_rows:
            g = "auction" if QuotationType.is_auction(int(t)) else "nego"
            grp[g]["count"] += int(cnt or 0)
            grp[g]["awarded"] += int(awarded or 0)

        sav = {"nego": [], "auction": []}
        for r in win_rows:
            g = "auction" if QuotationType.is_auction(int(r.type)) else "nego"
            sav[g].append(int(r.target_price) - int(r.bid_price))

        out = []
        for g, label in (("nego", "협상 (1:1)"), ("auction", "견적 (1:N)")):
            cnt = grp[g]["count"]
            rate = (grp[g]["awarded"] / cnt) if cnt else 0.0
            avg = int(sum(sav[g]) / len(sav[g])) if sav[g] else 0
            out.append(StatTypeRow(label=label, award_rate=rate, avg_savings=avg, count=cnt))
        return out

    def _cards(self, card_rows, drops: dict) -> list:
        by = {int(ct): int(n) for ct, n in card_rows if ct is not None}
        return [
            StatCardUsage(type="nego", label="협상카드", uses=by.get(CardType.NEGO.value, 0), avg_drop=int(round(drops.get(CardType.NEGO.value, 0)))),
            StatCardUsage(type="wild", label="와일드카드", uses=by.get(CardType.WILD.value, 0), avg_drop=int(round(drops.get(CardType.WILD.value, 0)))),
        ]

    def _card_drops(self, rows) -> dict:
        # 카드 사용 직후 제시가 하락(유형별 평균).
        #  - 일반: 카드 직전 유저 제시가 − 직후 유저 제시가.
        #  - 1% 인하(수락은 가격 재입력이 아님): 카드 직전 유저 제시가 − 최종 낙찰가(타결 세션).
        # rows: (session_id, seq, sender, target_price, card_used_yn, card_type, is_1pct, bid_price, status)
        by_sess: dict = {}
        for r in rows:
            by_sess.setdefault(r[0], []).append(r)
        sums = {CardType.NEGO.value: 0, CardType.WILD.value: 0}
        cnts = {CardType.NEGO.value: 0, CardType.WILD.value: 0}
        for chs in by_sess.values():
            for i, ch in enumerate(chs):
                _sid, _seq, _sender, _tp, used, ctype, is_1pct, bid_price, status = ch
                if not (used or is_1pct):
                    continue
                ct = int(ctype) if ctype else (CardType.WILD.value if is_1pct else None)
                if ct not in sums:
                    continue
                prev = next((c[3] for c in reversed(chs[:i]) if c[2] == ChatSender.USER.value and c[3] and c[3] > 0), None)
                if is_1pct:
                    if prev is not None and bid_price and status == SessionStatus.DONE.value and prev >= bid_price:
                        sums[ct] += (prev - bid_price)
                        cnts[ct] += 1
                else:
                    after = next((c[3] for c in chs[i + 1:] if c[2] == ChatSender.USER.value and c[3] and c[3] > 0), None)
                    if prev is not None and after is not None:
                        sums[ct] += (prev - after)
                        cnts[ct] += 1
        return {ct: (sums[ct] / cnts[ct]) if cnts[ct] else 0 for ct in sums}

    # ── 창(최근 6개월) ─────────────────────────────────────────
    def _window(self, now):
        yy, mm = now.year, now.month
        mm -= (WINDOW_MONTHS - 1)
        while mm <= 0:
            mm += 12
            yy -= 1
        window_start = now.replace(year=yy, month=mm, day=1, hour=0, minute=0, second=0, microsecond=0)
        labels, ly, lm = [], yy, mm
        for _ in range(WINDOW_MONTHS):
            labels.append(f"{ly:04d}-{lm:02d}")
            lm += 1
            if lm > 12:
                lm = 1
                ly += 1
        return labels, window_start

    # ── DB 실행 헬퍼 ───────────────────────────────────────────
    async def _read(self, fn) -> list:
        err, rows = await DB_SESSION_MNG.execute_lambda(quotations.DBType(), DBWRType.DB_READ.value, fn)
        return rows if err == ErrorType.SUCCESS else []

    async def _read_scalar(self, fn) -> float:
        err, val = await DB_SESSION_MNG.execute_lambda(quotations.DBType(), DBWRType.DB_READ.value, fn)
        return val if err == ErrorType.SUCCESS else 0.0