trait_engine.py

"""Trait 벡터 계산 + 유형 분류 + 매칭 엔진"""

import json
import math
import os
from scenarios import ALL_AXES, DATING_TYPES


def empty_vector() -> dict:
    """15축 0 벡터 생성"""
    return {axis: 0.0 for axis in ALL_AXES}


def apply_effects(vector: dict, effects: dict, multiplier: float = 1.0) -> dict:
    """trait_effect를 벡터에 적용"""
    for axis, delta in effects.items():
        if axis in vector:
            vector[axis] = max(-5.0, min(5.0, vector[axis] + delta * multiplier))
    return vector


def classify_type(vector: dict) -> tuple[str, dict[str, float]]:
    """유형 카드 분류 — 가중 점수 계산 후 Top-1 반환"""
    scores = {}
    for type_id, dtype in DATING_TYPES.items():
        score = 0.0
        for axis, weight in dtype["weights"].items():
            score += vector.get(axis, 0.0) * weight
        scores[type_id] = round(score, 2)

    best = max(scores, key=scores.get)
    return best, scores


def vector_magnitude(v: dict) -> float:
    return math.sqrt(sum(val ** 2 for val in v.values()))


def cosine_similarity(a: dict, b: dict) -> float:
    """두 벡터의 코사인 유사도"""
    axes = set(a.keys()) & set(b.keys())
    dot = sum(a[k] * b[k] for k in axes)
    mag_a = math.sqrt(sum(a[k] ** 2 for k in axes))
    mag_b = math.sqrt(sum(b[k] ** 2 for k in axes))
    if mag_a == 0 or mag_b == 0:
        return 0.0
    return dot / (mag_a * mag_b)


def merge_vectors(cumulative: dict, current: dict, weight_cum: float = 0.6) -> dict:
    """누적 벡터 + 현재 런 벡터 병합"""
    merged = {}
    weight_cur = 1.0 - weight_cum
    for axis in ALL_AXES:
        merged[axis] = round(
            cumulative.get(axis, 0.0) * weight_cum + current.get(axis, 0.0) * weight_cur,
            3,
        )
    return merged


def vector_to_profile(vector: dict, type_id: str) -> dict:
    """15축 벡터를 seed_profiles.json과 동일한 포맷으로 변환"""
    from scenarios import BEHAVIOR_AXES, BIG5_AXES
    dt = DATING_TYPES.get(type_id, {})
    return {
        "archetype": dt.get("name", ""),
        "traits": {axis: round(vector.get(axis, 0.0), 2) for axis in BEHAVIOR_AXES},
        "big_five": {axis: round(vector.get(axis, 0.0), 2) for axis in BIG5_AXES},
    }


def format_profile_display(profile: dict) -> str:
    """seed_profiles 포맷 프로필을 CLI용으로 출력"""
    lines = []
    lines.append(f"  아키타입: {profile.get('archetype', '?')}")
    lines.append("")
    lines.append("  ── 행동 성향 (10축) ──")
    for axis, val in profile.get("traits", {}).items():
        bar_len = min(int(abs(val) * 4), 20)
        if val >= 0:
            bar = "█" * bar_len + "░" * (20 - bar_len)
        else:
            bar = "░" * (20 - bar_len) + "█" * bar_len
        lines.append(f"    {axis:>22s}  {bar} {val:+.2f}")
    lines.append("")
    lines.append("  ── 성격 (Big Five) ──")
    for axis, val in profile.get("big_five", {}).items():
        bar_len = min(int(abs(val) * 4), 20)
        if val >= 0:
            bar = "█" * bar_len + "░" * (20 - bar_len)
        else:
            bar = "░" * (20 - bar_len) + "█" * bar_len
        lines.append(f"    {axis:>22s}  {bar} {val:+.2f}")
    return "\n".join(lines)


def format_profile_md(profile: dict) -> str:
    """seed_profiles 포맷 프로필을 마크다운으로 출력"""
    md = f"**아키타입:** {profile.get('archetype', '?')}\n\n"
    md += "#### 행동 성향 (10축)\n"
    md += "| 축 | 값 |\n|---|---|\n"
    for axis, val in profile.get("traits", {}).items():
        md += f"| {axis} | `{val:+.2f}` |\n"
    md += "\n#### 성격 (Big Five)\n"
    md += "| 축 | 값 |\n|---|---|\n"
    for axis, val in profile.get("big_five", {}).items():
        md += f"| {axis} | `{val:+.2f}` |\n"
    return md


def format_profile_json(profile: dict) -> str:
    """seed_profiles 포맷 프로필을 JSON 문자열로"""
    return json.dumps(profile, ensure_ascii=False, indent=2)


def format_type_card(type_id: str) -> str:
    """유형 카드 출력"""
    dt = DATING_TYPES[type_id]
    return f"""
╔══════════════════════════════════════╗
║  {dt['emoji']}  {dt['title']}
║
║  {dt['desc']}
╚══════════════════════════════════════╝
"""


# ── 시드 프로필 매칭 ────────────────────────────────────────

SEED_DATA_PATH = os.path.join(os.path.dirname(__file__), "data", "seed_profiles_1000.json")


def load_seed_profiles() -> list[dict]:
    """seed_profiles.json 로드"""
    with open(SEED_DATA_PATH, encoding="utf-8") as f:
        return json.load(f)


def profile_to_vector(profile: dict) -> dict:
    """시드 프로필의 traits + big_five를 15축 벡터로 변환"""
    vector = {}
    vector.update(profile.get("traits", {}))
    vector.update(profile.get("big_five", {}))
    return vector


def find_matches(user_vector: dict, user_type: str, top_n: int = 3) -> list[dict]:
    """시드 프로필 80명 중 매칭 상위 N명 추천

    match_score = 0.4 * behavior_score + 0.3 * big5_score + 0.2 * cluster_score + 0.1 * diversity
    """
    profiles = load_seed_profiles()
    results = []

    for profile in profiles:
        candidate_vector = profile_to_vector(profile)

        # 행동 벡터 유사도
        behavior_score = cosine_similarity(
            {k: user_vector.get(k, 0) for k in ALL_AXES[:10]},
            {k: candidate_vector.get(k, 0) for k in ALL_AXES[:10]},
        )

        # Big Five 유사도
        big5_score = cosine_similarity(
            {k: user_vector.get(k, 0) for k in ALL_AXES[10:]},
            {k: candidate_vector.get(k, 0) for k in ALL_AXES[10:]},
        )

        # 클러스터 점수 (같은 유형이면 보너스, 보완 유형이면 더 높은 보너스)
        cand_type, _ = classify_type(candidate_vector)
        complementary = {
            "flame": "stable", "stable": "flame",
            "emotional": "strategic", "strategic": "emotional",
            "free": "stable",
        }
        if cand_type == complementary.get(user_type, ""):
            cluster_score = 1.0
        elif cand_type == user_type:
            cluster_score = 0.7
        else:
            cluster_score = 0.3

        match_score = (
            0.4 * behavior_score
            + 0.3 * big5_score
            + 0.2 * cluster_score
            + 0.1 * 0.5  # 기본 다양성 점수
        )

        results.append({
            "profile": profile,
            "type": cand_type,
            "match_score": round(match_score * 100, 1),
            "behavior_sim": round(behavior_score * 100, 1),
            "big5_sim": round(big5_score * 100, 1),
        })

    results.sort(key=lambda x: x["match_score"], reverse=True)
    return results[:top_n]


AXIS_NAMES_KR = {
    "cooperation": "협력성", "leadership": "주도성", "emotional_depth": "감정 깊이",
    "pace": "관계 속도", "humor": "유머", "risk": "모험성",
    "contact_frequency": "연락 빈도", "affection": "애정 표현",
    "jealousy": "질투/독점", "planning": "계획성",
    "openness": "개방성", "conscientiousness": "성실성",
    "extraversion": "외향성", "agreeableness": "친화성", "neuroticism": "정서 불안정성",
}

COMPLEMENTARY_TYPES = {
    "flame": "stable", "stable": "flame",
    "emotional": "strategic", "strategic": "emotional",
    "free": "stable",
}


def generate_match_reason(user_vector: dict, user_type: str, match: dict) -> str:
    """매칭 이유를 deterministic하게 생성."""
    cand_vector = profile_to_vector(match["profile"])

    similarities = []
    complements = []
    for axis in ALL_AXES:
        u = user_vector.get(axis, 0)
        c = cand_vector.get(axis, 0)
        diff = abs(u - c)
        magnitude = abs(u) + abs(c)
        if magnitude > 1.0:
            if diff < 1.0:
                similarities.append((axis, magnitude))
            elif diff > 1.5 and u * c < 0:
                complements.append((axis, abs(u), abs(c)))

    similarities.sort(key=lambda x: -x[1])
    parts = []

    if similarities:
        names = [AXIS_NAMES_KR.get(s[0], s[0]) for s in similarities[:2]]
        parts.append(f"{', '.join(names)}이(가) 비슷해요")

    if complements:
        name = AXIS_NAMES_KR.get(complements[0][0], complements[0][0])
        parts.append(f"{name}에서 서로 보완돼요")

    cand_type = match.get("type", "")
    if cand_type == COMPLEMENTARY_TYPES.get(user_type, ""):
        parts.append("상호 보완적인 유형이에요")
    elif cand_type == user_type:
        parts.append("같은 유형이라 공감대가 높아요")

    return " · ".join(parts) if parts else "전반적인 성향이 잘 맞아요"


def format_match_card(match: dict, rank: int) -> str:
    """매칭 결과 카드 출력"""
    p = match["profile"]
    dt = DATING_TYPES.get(match["type"], {})
    return f"""\
  ┌──────────────────────────────────────┐
  │  #{rank}  💞 {p['name']} ({p['age']}세, {p['gender']})
  │  유형: {dt.get('emoji', '')} {dt.get('name', '')}  ({p.get('archetype', '')})
  │
  │  매칭 적합도  {match['match_score']}%
  │  행동 유사도  {match['behavior_sim']}%
  │  성격 유사도  {match['big5_sim']}%
  └──────────────────────────────────────┘"""