medical_training/apps/user/stats.py

"""移动端个人中心 - 训练统计 / 智能分析（读 fastapi 只读训练表）。

- 4.3 临床核心能力指标   GET /api/user/competency-metrics/
- 4.4 训练记录（统计信息） GET /api/user/training-records/
- 4.5 智能分析（关联评价表）GET /api/user/analysis/

数据源：training_record（managed=False 只读，user_id = 当前用户.id）。
各能力维度得分 + 满分来自 training_record.ai_feedback_structured.dimension_scores
（与 training_score_detail 同源，但 json 里带 max_score，便于算「得分率」）。
"""
import json
from datetime import timedelta

from django.db.models import Sum, Avg, Case, When, F, DecimalField
from django.utils import timezone
from rest_framework.decorators import api_view, permission_classes
from rest_framework.permissions import IsAuthenticated
from rest_framework.pagination import PageNumberPagination
from rest_framework.response import Response
from drf_spectacular.utils import extend_schema

from apps.training.models import TrainingRecord

# 临床胜任力标准 5 维（与 training_score_detail / ai_feedback_structured 的评分维度一致）
STANDARD_DIMS = ['信息获取', '分析推理', '处置决策', '沟通人文', '临床整合']

# 评分维度（fastapi 实际打分维度，名称随病例评分规则而变）→ 标准 5 维 的归并映射。
# 维度名不完全统一（实测有 检查利用/临床推理/人文沟通 等同义写法），统一归并到 5 个标准维度。
# 未在映射内的维度会被忽略。
DIMENSION_MAP = {
    # 信息获取（病史/问诊）
    '信息获取': '信息获取', '病史采集': '信息获取', '问诊': '信息获取',
    # 分析推理（诊断/鉴别/推理）
    '分析推理': '分析推理', '临床推理': '分析推理', '诊断推理': '分析推理',
    '鉴别诊断': '分析推理', '诊断能力': '分析推理',
    # 处置决策（检查决策 + 治疗决策合并）
    '处置决策': '处置决策', '检查利用': '处置决策', '检查理解': '处置决策',
    '检查决策': '处置决策', '辅助检查': '处置决策', '检验决策': '处置决策',
    '治疗决策': '处置决策', '治疗': '处置决策',
    # 沟通人文
    '沟通人文': '沟通人文', '人文沟通': '沟通人文', '沟通技巧': '沟通人文',
    '医患沟通': '沟通人文', '沟通': '沟通人文',
    # 临床整合（含知识掌握/运用）
    '临床整合': '临床整合', '知识掌握': '临床整合', '知识运用': '临床整合',
}
# 诊断相关维度（诊断准确率口径：这些维度的平均得分率；与归并到「分析推理」的来源一致）
DIAGNOSIS_DIMS = {'诊断推理', '诊断能力', '鉴别诊断', '分析推理', '临床推理'}


def _completed_qs(user):
    return TrainingRecord.objects.filter(user_id=user.id, status='completed')


# total_score 归一到百分制：五分制(0-5) ×20，其余按原值(percentage)。
# 用于 avg_score / current_score / 趋势 / 较上周，避免百分制与五分制混平均失真。
# （诊断准确率/雷达用各维度 score/max_score 比值自归一，不走这里。）
def _norm_total(score, score_type):
    if score is None:
        return None
    return float(score) * 20 if score_type == 'five_point' else float(score)


# DB 侧等价表达式，供 aggregate(Avg(...)) 使用
_NORM_TOTAL_EXPR = Case(
    When(score_type='five_point', then=F('total_score') * 20),
    default=F('total_score'),
    output_field=DecimalField(max_digits=7, decimal_places=2),
)


def _week_start(offset_weeks=0):
    now = timezone.localtime()
    monday = now - timedelta(days=now.weekday(), weeks=-offset_weeks)
    return monday.replace(hour=0, minute=0, second=0, microsecond=0)


def _dimension_scores(rec):
    """从 ai_feedback_structured.dimension_scores 取 [(dimension, score, max_score)]。"""
    fb = rec.ai_feedback_structured
    if isinstance(fb, str):
        try:
            fb = json.loads(fb)
        except Exception:
            fb = {}
    if not isinstance(fb, dict):
        return []
    out = []
    for d in (fb.get('dimension_scores') or []):
        if isinstance(d, dict) and d.get('dimension'):
            out.append((d['dimension'], d.get('score'), d.get('max_score')))
    return out


def _diagnosis_accuracy(records):
    """诊断相关维度的平均得分率（%）。无数据返回 None。"""
    rates = []
    for rec in records:
        for dim, score, mx in _dimension_scores(rec):
            if dim in DIAGNOSIS_DIMS and score is not None and mx and float(mx) > 0:
                rates.append(float(score) / float(mx))
    return round(sum(rates) / len(rates) * 100) if rates else None


def _hours(seconds):
    return round((seconds or 0) / 3600, 1)


# ── 4.3 临床核心能力指标 ─────────────────────────────────────────────────────

@extend_schema(summary='临床核心能力指标', tags=['个人中心'])
@api_view(['GET'])
@permission_classes([IsAuthenticated])
def competency_metrics(request):
    qs = _completed_qs(request.user)
    agg = qs.aggregate(total=Sum('duration_seconds'), avg=Avg(_NORM_TOTAL_EXPR))
    avg = agg['avg']
    return Response({
        'completed_cases': qs.count(),
        'completed_cases_week': qs.filter(end_time__gte=_week_start()).count(),
        'total_hours': _hours(agg['total']),
        'avg_score': round(float(avg), 1) if avg is not None else 0,
        'diagnosis_accuracy': _diagnosis_accuracy(qs.only('ai_feedback_structured')),
    })


# ── 4.4 训练记录（统计信息）─────────────────────────────────────────────────

@extend_schema(summary='训练记录（统计信息）', tags=['个人中心'])
@api_view(['GET'])
@permission_classes([IsAuthenticated])
def training_records(request):
    base = _completed_qs(request.user)
    search = (request.query_params.get('search') or '').strip()
    if search:
        from django.db.models import Q
        base = base.filter(Q(case__title__icontains=search) | Q(case__department__name__icontains=search))

    agg = base.aggregate(total=Sum('duration_seconds'))
    summary = {
        'total_cases': base.count(),
        'total_hours': _hours(agg['total']),
        'avg_accuracy': _diagnosis_accuracy(base.only('ai_feedback_structured')),
    }

    page_qs = base.select_related('case', 'case__department').order_by('-end_time', '-id')
    paginator = PageNumberPagination()
    page = paginator.paginate_queryset(page_qs, request)
    results = []
    for r in page:
        case = r.case if r.case_id else None        # 孤立外键(本地无对应病例)时为 None
        dept = case.department if case else None     # 病例无科室或科室缺失时为 None
        results.append({
            'record_id': r.id,
            'case_id': r.case_id,
            'case_title': case.title if case else '',
            'department': dept.name if dept else '',
            'trained_at': r.end_time or r.created_at,
            'score': float(r.total_score) if r.total_score is not None else None,
            'score_type': r.score_type,
            'evaluation_level': r.evaluation_level,
            'training_mode': r.training_mode,
            'case_type': r.case_type,
        })
    resp = paginator.get_paginated_response(results)
    resp.data['summary'] = summary
    return resp


# ── 4.5 智能分析（关联评价表）─────────────────────────────────────────────────

@extend_schema(summary='智能分析（关联评价表）', tags=['个人中心'])
@api_view(['GET'])
@permission_classes([IsAuthenticated])
def analysis(request):
    recs = list(_completed_qs(request.user).order_by('end_time', 'id'))

    # 趋势 / 当前评分（按 total_score，按 score_type 归一到百分制）
    scored = [(r.end_time or r.created_at, _norm_total(r.total_score, r.score_type))
              for r in recs if r.total_score is not None]
    recent = scored[-7:]
    recent_trend = [{'label': (dt.strftime('%m-%d') if dt else ''), 'score': round(s)} for dt, s in recent]
    current_score = round(sum(s for _, s in recent) / len(recent), 1) if recent else 0

    # 较上周提升%
    ws, ws_prev = _week_start(), _week_start(offset_weeks=-1)
    this_week = [s for dt, s in scored if dt and dt >= ws]
    last_week = [s for dt, s in scored if dt and ws_prev <= dt < ws]
    score_delta_pct = None
    if this_week and last_week:
        a, b = sum(this_week) / len(this_week), sum(last_week) / len(last_week)
        if b:
            score_delta_pct = round((a - b) / b * 100)

    # 雷达：维度得分率归一到 0-100，按 A 组归并求均
    from collections import defaultdict
    buckets = defaultdict(list)
    for r in recs:
        for dim, score, mx in _dimension_scores(r):
            std = DIMENSION_MAP.get(dim)
            if std and score is not None and mx and float(mx) > 0:
                buckets[std].append(float(score) / float(mx) * 100)
    radar = [{'dimension': d, 'score': (round(sum(buckets[d]) / len(buckets[d])) if buckets[d] else 0)}
             for d in STANDARD_DIMS]
    # 仅在「有数据」的维度里排序取强/弱，避免无数据维度(0 分)被误判为薄弱项
    ordered = sorted((x for x in radar if buckets[x['dimension']]), key=lambda x: x['score'])
    weak_dimensions = [ordered[0]['dimension']] if ordered else []

    comment = ''
    if len(ordered) >= 2 and ordered[0]['score'] != ordered[-1]['score']:
        # 强弱维度不同才做对比，避免「最强==最弱」时出现自相矛盾文案
        comment = f"您的{ordered[-1]['dimension']}表现突出，但{ordered[0]['dimension']}仍有提升空间，建议增加相关模拟训练。"
    elif ordered:
        comment = f"各维度表现较为均衡，可针对{weak_dimensions[0]}做进一步强化。"

    return Response({
        'current_score': current_score,
        'score_delta_pct': score_delta_pct,
        'recent_trend': recent_trend,
        'radar': radar,
        'weak_dimensions': weak_dimensions,
        'comment': comment,
    })