Merge pull request 'fix: prediction e2e — Decimal/violation/KPI/stats/parent workflow 5건 수정' (#12) from feature/prediction-e2e-fixes into develop

backend/src/main/resources/db/migration/V016__parent_workflow_columns.sql

@@ -0,0 +1,64 @@
+-- V016: parent workflow 관련 테이블에 prediction 코드가 요구하는 누락 컬럼 추가
+-- gear_parent_inference.py, gear_parent_episode.py가 참조하는 컬럼들
+
+-- === gear_parent_label_sessions ===
+ALTER TABLE kcg.gear_parent_label_sessions
+    ADD COLUMN IF NOT EXISTS label_parent_name VARCHAR(200),
+    ADD COLUMN IF NOT EXISTS label_parent_vessel_id BIGINT,
+    ADD COLUMN IF NOT EXISTS normalized_parent_name VARCHAR(100),
+    ADD COLUMN IF NOT EXISTS duration_days INT DEFAULT 3,
+    ADD COLUMN IF NOT EXISTS actor VARCHAR(100),
+    ADD COLUMN IF NOT EXISTS comment TEXT,
+    ADD COLUMN IF NOT EXISTS metadata JSONB,
+    ADD COLUMN IF NOT EXISTS updated_at TIMESTAMPTZ DEFAULT now();
+
+CREATE INDEX IF NOT EXISTS idx_label_session_norm
+    ON kcg.gear_parent_label_sessions(normalized_parent_name, active_from DESC)
+    WHERE status = 'ACTIVE';
+
+-- === gear_group_parent_resolution ===
+ALTER TABLE kcg.gear_group_parent_resolution
+    ADD COLUMN IF NOT EXISTS parent_name VARCHAR(200),
+    ADD COLUMN IF NOT EXISTS normalized_parent_name VARCHAR(100),
+    ADD COLUMN IF NOT EXISTS selected_parent_name VARCHAR(200),
+    ADD COLUMN IF NOT EXISTS selected_vessel_id BIGINT,
+    ADD COLUMN IF NOT EXISTS confidence NUMERIC(7,4),
+    ADD COLUMN IF NOT EXISTS decision_source VARCHAR(30),
+    ADD COLUMN IF NOT EXISTS top_score NUMERIC(7,4),
+    ADD COLUMN IF NOT EXISTS second_score NUMERIC(7,4),
+    ADD COLUMN IF NOT EXISTS score_margin NUMERIC(7,4),
+    ADD COLUMN IF NOT EXISTS stable_cycles INT DEFAULT 0,
+    ADD COLUMN IF NOT EXISTS evidence_summary JSONB,
+    ADD COLUMN IF NOT EXISTS episode_id VARCHAR(50),
+    ADD COLUMN IF NOT EXISTS continuity_source VARCHAR(30),
+    ADD COLUMN IF NOT EXISTS continuity_score NUMERIC(7,4),
+    ADD COLUMN IF NOT EXISTS prior_bonus_total NUMERIC(7,4);
+
+CREATE INDEX IF NOT EXISTS idx_parent_resolution_episode
+    ON kcg.gear_group_parent_resolution(episode_id);
+
+-- === gear_parent_candidate_exclusions ===
+-- gear_parent_inference.py가 참조하는 추가 컬럼
+-- 참고: 코드는 candidate_mmsi를 쿼리 — excluded_mmsi의 generated column으로 매핑
+ALTER TABLE kcg.gear_parent_candidate_exclusions
+    ADD COLUMN IF NOT EXISTS normalized_parent_name VARCHAR(100),
+    ADD COLUMN IF NOT EXISTS reason_type VARCHAR(50),
+    ADD COLUMN IF NOT EXISTS duration_days INT,
+    ADD COLUMN IF NOT EXISTS metadata JSONB,
+    ADD COLUMN IF NOT EXISTS updated_at TIMESTAMPTZ DEFAULT now(),
+    ADD COLUMN IF NOT EXISTS active_from TIMESTAMPTZ DEFAULT now(),
+    ADD COLUMN IF NOT EXISTS active_until TIMESTAMPTZ;
+
+-- candidate_mmsi 별칭 (generated column)
+ALTER TABLE kcg.gear_parent_candidate_exclusions
+    ADD COLUMN IF NOT EXISTS candidate_mmsi VARCHAR(20)
+    GENERATED ALWAYS AS (excluded_mmsi) STORED;
+
+-- === gear_group_parent_resolution 추가 타임스탬프 컬럼 ===
+ALTER TABLE kcg.gear_group_parent_resolution
+    ADD COLUMN IF NOT EXISTS last_evaluated_at TIMESTAMPTZ,
+    ADD COLUMN IF NOT EXISTS last_promoted_at TIMESTAMPTZ;
+
+-- === gear_group_parent_candidate_snapshots ===
+ALTER TABLE kcg.gear_group_parent_candidate_snapshots
+    ADD COLUMN IF NOT EXISTS parent_name VARCHAR(200);

prediction/algorithms/gear_correlation.py

@@ -781,8 +781,9 @@ def _load_all_scores(conn) -> dict[tuple, dict]:
         result = {}
         for row in cur.fetchall():
             key = (row[0], row[1], row[2], row[3])
+            # psycopg2가 NUMERIC을 Decimal로 반환하므로 float으로 변환 (float 상수와의 연산 호환)
             result[key] = {
-                'current_score': row[4],
+                'current_score': float(row[4]) if row[4] is not None else 0.0,
                 'streak_count': row[5],
                 'last_observed_at': row[6],
                 'target_type': row[7],

prediction/output/kpi_writer.py

@@ -4,7 +4,7 @@
 매 분석 사이클마다 오늘 날짜 기준 카운트를 계산하여 6개 KPI 갱신.
 """
 import logging
-from datetime import date, datetime, timezone
+from datetime import datetime, timedelta, timezone
 
 from config import qualified_table
 from db.kcgdb import get_conn
@@ -16,16 +16,20 @@ EVENTS_TABLE = qualified_table('prediction_events')
 ENF_TABLE = qualified_table('enforcement_records')
 VAR_TABLE = qualified_table('vessel_analysis_results')
 
+# 한국 표준시 (운영 기준)
+_KST = timezone(timedelta(hours=9))
+
 
 def run_kpi_writer() -> dict:
     """
-    오늘 날짜 기준으로 6개 KPI를 재계산하여 갱신.
+    오늘(KST) 날짜 기준으로 6개 KPI를 재계산하여 갱신.
 
     Returns:
         { kpi_key: value } 딕셔너리
     """
-    today = date.today()
+    # KST 기준 "오늘" 시작 시각 (해당 시각은 UTC로도 비교 가능하므로 DB 필드가 TIMESTAMPTZ면 안전)
-    today_start = datetime(today.year, today.month, today.day, tzinfo=timezone.utc)
+    now_kst = datetime.now(_KST)
+    today_start = now_kst.replace(hour=0, minute=0, second=0, microsecond=0)
     now = datetime.now(timezone.utc)
     results = {}
 

prediction/output/stats_aggregator.py

@@ -22,6 +22,9 @@ VAR_TABLE = qualified_table('vessel_analysis_results')
 EVENTS_TABLE = qualified_table('prediction_events')
 ENF_TABLE = qualified_table('enforcement_records')
 
+# 한국 표준시 (운영 기준 — 일/월 집계 경계)
+_KST = timezone(timedelta(hours=9))
+
 
 def _jsonb(d: dict) -> str:
     return json.dumps(d, ensure_ascii=False)
@@ -92,9 +95,10 @@ def aggregate_hourly(target_hour: Optional[datetime] = None) -> dict:
 
 
 def aggregate_daily(target_date: Optional[date] = None) -> dict:
-    """지정 날짜 기준 daily 집계."""
+    """지정 날짜 기준 daily 집계 (KST 기준)."""
-    d = target_date or date.today()
+    d = target_date or datetime.now(_KST).date()
-    day_start = datetime(d.year, d.month, d.day, tzinfo=timezone.utc)
+    # KST 자정을 TIMESTAMPTZ로 표현 (UTC -9시간)
+    day_start = datetime(d.year, d.month, d.day, tzinfo=_KST)
     day_end = day_start + timedelta(days=1)
     now = datetime.now(timezone.utc)
 
@@ -186,8 +190,8 @@ def aggregate_daily(target_date: Optional[date] = None) -> dict:
 
 
 def aggregate_monthly(target_month: Optional[date] = None) -> dict:
-    """지정 월 기준 monthly 집계 (daily 합산)."""
+    """지정 월 기준 monthly 집계 (daily 합산, KST 기준)."""
-    d = target_month or date.today().replace(day=1)
+    d = target_month or datetime.now(_KST).date().replace(day=1)
     month_start = d.replace(day=1)
     if month_start.month == 12:
         month_end = month_start.replace(year=month_start.year + 1, month=1)

prediction/output/violation_classifier.py

@@ -16,7 +16,15 @@ VAR_TABLE = qualified_table('vessel_analysis_results')
 
 
 def classify_violations(result: dict) -> list[str]:
-    """단일 분석 결과에 대해 위반 유형 리스트 반환."""
+    """단일 분석 결과에 대해 위반 유형 리스트 반환.
+
+    판정 기준:
+    - EEZ_VIOLATION: 중국선박(412*) + EEZ/NLL/특별금어구역 + 비허가
+    - DARK_VESSEL: is_dark + 30분 이상 갭
+    - MMSI_TAMPERING: spoofing_score > 0.6
+    - ILLEGAL_TRANSSHIP: transship_suspect
+    - RISK_BEHAVIOR: 위반 없이 risk_score >= 70
+    """
     violations = []
 
     zone = result.get('zone_code', '') or ''
@@ -24,13 +32,18 @@ def classify_violations(result: dict) -> list[str]:
     is_dark = result.get('is_dark', False)
     spoofing = result.get('spoofing_score', 0) or 0
     transship = result.get('transship_suspect', False)
-    permit = result.get('permit_status', 'UNKNOWN') or 'UNKNOWN'
     gap_min = result.get('gap_duration_min', 0) or 0
+    mmsi = str(result.get('mmsi', '') or '')
+    # permit_status는 선택적 — 없으면 중국 선박인지로 판단 (412* prefix)
+    permit = result.get('permit_status') or ''
+    is_chinese = mmsi.startswith('412') or mmsi.startswith('413')
 
-    # EEZ 침범
+    # EEZ 침범: 중국선박이 한국 해역에 진입 (중국선박은 기본적으로 비허가 상정)
     if zone in ('NLL', 'SPECIAL_FISHING_1', 'SPECIAL_FISHING_2',
                 'SPECIAL_FISHING_3', 'SPECIAL_FISHING_4', 'EEZ_KR'):
-        if permit in ('NONE', 'EXPIRED', 'REVOKED'):
+        if is_chinese and permit not in ('VALID', 'PERMITTED'):
+            violations.append('EEZ_VIOLATION')
+        elif permit in ('NONE', 'EXPIRED', 'REVOKED'):
             violations.append('EEZ_VIOLATION')
 
     # 다크베셀
@@ -45,7 +58,7 @@ def classify_violations(result: dict) -> list[str]:
     if transship:
         violations.append('ILLEGAL_TRANSSHIP')
 
-    # 어구 불법 (gear_judgment이 있는 경우)
+    # 어구 불법 (gear_judgment이 있는 경우만 — 현재는 scheduler에서 채우지 않음)
     gear_judgment = result.get('gear_judgment', '') or ''
     if gear_judgment in ('NO_PERMIT', 'GEAR_MISMATCH', 'ZONE_VIOLATION', 'SEASON_VIOLATION'):
         violations.append('ILLEGAL_GEAR')
@@ -61,6 +74,8 @@ def run_violation_classifier(analysis_results: list[dict]) -> dict:
     """
     분석 결과 리스트에 위반 카테고리를 라벨링하고 DB 업데이트.
 
+    AnalysisResult에는 DB id가 없으므로 (mmsi, analyzed_at)으로 UPDATE.
+
     Returns:
         { 'classified': int, 'violations_found': int }
     """
@@ -69,16 +84,18 @@ def run_violation_classifier(analysis_results: list[dict]) -> dict:
 
     for result in analysis_results:
         violations = classify_violations(result)
-        result_id = result.get('id')
+        mmsi = result.get('mmsi')
-        if result_id and violations:
+        analyzed_at = result.get('analyzed_at')
-            updates.append((violations, result_id))
+        if mmsi and analyzed_at and violations:
+            updates.append((violations, str(mmsi), analyzed_at))
             violations_found += len(violations)
 
     if updates:
         with get_conn() as conn:
             execute_batch(
                 conn.cursor(),
-                f"UPDATE {VAR_TABLE} SET violation_categories = %s WHERE id = %s",
+                f"UPDATE {VAR_TABLE} SET violation_categories = %s "
+                f"WHERE mmsi = %s AND analyzed_at = %s",
                 updates,
             )
             conn.commit()