perf: filter scan optimization and portfolio selection improvements

Precompute p_ball to speed up exhaustive filtering, add fixed-ball validation with labeled exceptions, and improve portfolio selection via ymd-seeded shuffle and coverage-aware tie-breaking. Include lotto draw 1225 history update.

Co-authored-by: Cursor <cursoragent@cursor.com>

final_BallFilter.py

@@ -3812,7 +3812,8 @@ class BallFilter:
         if len(set_ball & {3, 20, 44}) == 3: return 2928
         return None
 
-    def extract_final_candidates(self, ball, no=None, until_end=False, df=None):
+    def extract_final_candidates(self, ball, no=None, until_end=False, df=None, p_ball=None):
+        if p_ball is None:
             p_ball = df[df['no'] == no - 1].values.tolist()[0]
             p_ball = p_ball[1:7]
 
@@ -4447,7 +4448,9 @@ class BallFilter:
 
         return filter_set
 
-    def filter(self, ball, no, until_end=False, df=None, filter_ball=None):
-        filter_type = self.extract_final_candidates(ball=ball, no=no, until_end=until_end, df=df)
+    def filter(self, ball, no, until_end=False, df=None, filter_ball=None, p_ball=None):
+        filter_type = self.extract_final_candidates(
+            ball=ball, no=no, until_end=until_end, df=df, p_ball=p_ball
+        )
 
         return filter_type

final_practice.py

@@ -5,8 +5,10 @@ from DataCrawler import DataCrawler
 
 import json
 import os
+import random
 import pandas as pd
 import itertools
+from collections import Counter
 from datetime import datetime, timedelta
 from TelegramBot import TelegramBot
 
@@ -98,6 +100,70 @@ class Practice:
 
         return
 
+    def validate_fixed_balls(self, resources_path, ymd, fixed_balls):
+        """
+        고정수 BallFilter 통과 여부를 검증한다.
+
+        Returns:
+            dict: total, passed_count, failed_count, draw_no, details
+        """
+        lotto_history_json = os.path.join(resources_path, 'lotto_history.json')
+        ball_filter = BallFilter(lotto_history_json)
+        draw_no = ball_filter.getNextNo(ymd)
+
+        lotto_history_txt = os.path.join(resources_path, 'lotto_history.txt')
+        df_ball = pd.read_csv(lotto_history_txt, header=None)
+        df_ball.columns = ['no', 'b1', 'b2', 'b3', 'b4', 'b5', 'b6', 'bn']
+        prev_row = df_ball[df_ball['no'] == draw_no - 1].values.tolist()[0]
+        p_ball = prev_row[1:7]
+
+        details = []
+        passed_count = 0
+        for index, ball in enumerate(fixed_balls):
+            filter_type = ball_filter.filter(
+                ball=ball, no=draw_no, until_end=False, df=df_ball, p_ball=p_ball
+            )
+            passed = len(filter_type) == 0
+            if passed:
+                passed_count += 1
+            details.append({
+                'index': index + 1,
+                'ball': ball,
+                'passed': passed,
+                'filter_reasons': sorted(filter_type),
+            })
+
+        return {
+            'draw_no': draw_no,
+            'total': len(fixed_balls),
+            'passed_count': passed_count,
+            'failed_count': len(fixed_balls) - passed_count,
+            'details': details,
+        }
+
+    @staticmethod
+    def format_fixed_validation_summary(validation):
+        """고정수 검증 결과를 Telegram/로그용 문자열로 변환한다."""
+        lines = [
+            "   - 고정수 필터 검증: {}/{} 통과".format(
+                validation['passed_count'], validation['total']
+            )
+        ]
+        if validation['failed_count'] > 0:
+            lines.append(
+                "   - 필터 예외 포함: {}개 (고정수 유지)".format(
+                    validation['failed_count']
+                )
+            )
+            for item in validation['details']:
+                if item['passed']:
+                    continue
+                reason = item['filter_reasons'][0] if item['filter_reasons'] else 'unknown'
+                lines.append(
+                    "     * #{} {} -> {}".format(item['index'], item['ball'], reason)
+                )
+        return "\n".join(lines)
+
     def _can_add_ball(self, ball, fixed_balls, selected_balls, max_overlap):
         ball_set = set(ball)
 
@@ -111,11 +177,48 @@ class Practice:
 
         return True
 
-    def select_portfolio(self, fixed_balls, candidates, target_count):
+    @staticmethod
+    def _portfolio_number_counts(fixed_balls, selected_balls):
+        """포트폴리오 내 번호 등장 횟수를 집계한다."""
+        counts = Counter()
+        for ball in fixed_balls + selected_balls:
+            counts.update(ball)
+        return counts
+
+    @staticmethod
+    def _coverage_priority(ball, number_counts):
+        """낮을수록 포트폴리오에 덜 등장한 번호 위주 조합이다."""
+        return sum(number_counts.get(number, 0) for number in ball)
+
+    def _pick_best_candidate(self, unique_candidates, selected_keys, fixed_balls, selected, max_overlap):
+        """겹침 제약을 만족하는 후보 중 번호 커버리지가 가장 넓은 조합을 고른다."""
+        number_counts = self._portfolio_number_counts(fixed_balls, selected)
+        best_candidate = None
+        best_score = None
+        best_key = None
+
+        for candidate in unique_candidates:
+            key = tuple(candidate)
+            if key in selected_keys:
+                continue
+            if not self._can_add_ball(candidate, fixed_balls, selected, max_overlap):
+                continue
+
+            score = self._coverage_priority(candidate, number_counts)
+            if best_candidate is None or score < best_score or (score == best_score and key < best_key):
+                best_candidate = candidate
+                best_score = score
+                best_key = key
+
+        return best_candidate, best_key
+
+    def select_portfolio(self, fixed_balls, candidates, target_count, shuffle_seed=None):
         """
         2차 포트폴리오 선정:
         - 중복 제거
+        - shuffle_seed 기반 셔플로 순서 편향 완화
         - 고정수/선정수 간 중복도(겹치는 번호 수) 제약을 단계적으로 완화하며 선택
+        - 동률 후보는 번호 커버리지가 넓은 조합 우선
         """
         unique_candidates = []
         seen = set()
@@ -128,6 +231,10 @@ class Practice:
             seen.add(key)
             unique_candidates.append(list(key))
 
+        if shuffle_seed is not None:
+            rng = random.Random(int(shuffle_seed))
+            rng.shuffle(unique_candidates)
+
         if target_count <= 0:
             return []
 
@@ -139,26 +246,27 @@ class Practice:
         overlap_stages = [2, 3, 4, 5]
 
         for max_overlap in overlap_stages:
-            for candidate in unique_candidates:
-                key = tuple(candidate)
-                if key in selected_keys:
-                    continue
+            while len(selected) < target_count:
+                best_candidate, best_key = self._pick_best_candidate(
+                    unique_candidates, selected_keys, fixed_balls, selected, max_overlap
+                )
+                if best_candidate is None:
+                    break
+
+                selected.append(best_candidate)
+                selected_keys.add(best_key)
 
-                if self._can_add_ball(candidate, fixed_balls, selected, max_overlap):
-                    selected.append(candidate)
-                    selected_keys.add(key)
             if len(selected) >= target_count:
                 return selected
 
-        # 단계 완화 후에도 부족하면 남은 조합을 순서대로 채움
-        for candidate in unique_candidates:
-            key = tuple(candidate)
-            if key in selected_keys:
-                continue
-            selected.append(candidate)
-            selected_keys.add(key)
-            if len(selected) >= target_count:
+        while len(selected) < target_count:
+            best_candidate, best_key = self._pick_best_candidate(
+                unique_candidates, selected_keys, fixed_balls, selected, max_overlap=6
+            )
+            if best_candidate is None:
                 break
+            selected.append(best_candidate)
+            selected_keys.add(best_key)
 
         return selected
 
@@ -175,15 +283,19 @@ class Practice:
         df_ball = pd.read_csv(lottoHistoryFileName, header=None)
         df_ball.columns = ['no', 'b1', 'b2', 'b3', 'b4', 'b5', 'b6', 'bn']
 
+        prev_row = df_ball[df_ball['no'] == no - 1].values.tolist()[0]
+        p_ball = prev_row[1:7]
+
         passed_candidates = []
-        nCr = list(itertools.combinations(candidates, 6))
-        for idx, ball in enumerate(nCr):
+        for idx, ball in enumerate(itertools.combinations(candidates, 6)):
 
             if idx % 1000000 == 0:
                 print("   - {} processed, pass: {}".format(idx, len(passed_candidates)))
             ball = list(ball)
 
-            filter_type = ballFilter.filter(ball=ball, no=no, until_end=False, df=df_ball)
+            filter_type = ballFilter.filter(
+                ball=ball, no=no, until_end=False, df=df_ball, p_ball=p_ball
+            )
             filter_size = len(filter_type)
 
             if 0 < filter_size:
@@ -195,12 +307,11 @@ class Practice:
         selected_candidates = self.select_portfolio(
             fixed_balls=fixed_balls,
             candidates=passed_candidates,
-            target_count=variable_target_count
+            target_count=variable_target_count,
+            shuffle_seed=ymd,
         )
 
-        p_ball = df_ball[df_ball['no'] == no - 1].values.tolist()[0]
-        p_no = p_ball[0]
-        p_ball = p_ball[1:7]
+        p_no = prev_row[0]
 
         return p_no, p_ball, selected_candidates, len(passed_candidates), variable_target_count
 
@@ -243,6 +354,12 @@ if __name__ == '__main__':
     # 매주 고정
     fixed_balls = []
     practice.predict1(fixed_balls)
+    fixed_validation = practice.validate_fixed_balls(
+        resources_path=resources_path,
+        ymd=ymd,
+        fixed_balls=fixed_balls,
+    )
+    print(Practice.format_fixed_validation_summary(fixed_validation))
     result_json[ymd].extend(fixed_balls)
 
     # 필터 기반 예측
@@ -254,6 +371,15 @@ if __name__ == '__main__':
     )
     result_json[ymd].extend(selected_candidates)
 
+    if '_meta' not in result_json:
+        result_json['_meta'] = {}
+    result_json['_meta'][ymd] = {
+        'fixed_validation': fixed_validation,
+        'passed_count': passed_count,
+        'selected_count': len(selected_candidates),
+        'portfolio_shuffle_seed': ymd,
+    }
+
     with open(recommend_result_file, 'w', encoding='utf-8') as outFp:
         json.dump(result_json, outFp, ensure_ascii=False)
 
@@ -261,6 +387,7 @@ if __name__ == '__main__':
     total_cost = total_games * COST_PER_GAME
     p_str = "[지난주] {}\n   - {} 회차, {}\n[금주] {}\n   - {} 회차\n[모델#25]\n".format(last_weekend, p_no, str(p_ball), ymd, (p_no + 1))
     p_str += "   - 고정수: {}개\n".format(len(fixed_balls))
+    p_str += Practice.format_fixed_validation_summary(fixed_validation) + "\n"
     p_str += "   - 필터 통과 후보: {}개\n".format(passed_count)
     p_str += "   - 추가 선정: {}개 (목표 {}개)\n".format(len(selected_candidates), variable_target_count)
     p_str += "   - 총 추천: {}개, 총 금액: {:,}원 (한도 {:,}원)\n".format(total_games, total_cost, MAX_BUDGET_KRW)

resources/lotto_history.json

@@ -1222,3 +1222,4 @@
 {"returnValue": "success", "drwNoDate": "2026-05-02", "drwNo": 1222, "drwtNo1": 4, "drwtNo2": 11, "drwtNo3": 17, "drwtNo4": 22, "drwtNo5": 32, "drwtNo6": 41, "bnusNo": 34}
 {"returnValue": "success", "drwNoDate": "2026-05-09", "drwNo": 1223, "drwtNo1": 16, "drwtNo2": 18, "drwtNo3": 20, "drwtNo4": 32, "drwtNo5": 33, "drwtNo6": 39, "bnusNo": 26}
 {"returnValue": "success", "drwNoDate": "2026-05-16", "drwNo": 1224, "drwtNo1": 9, "drwtNo2": 18, "drwtNo3": 21, "drwtNo4": 27, "drwtNo5": 44, "drwtNo6": 45, "bnusNo": 28}
+{"returnValue": "success", "drwNoDate": "2026-05-23", "drwNo": 1225, "drwtNo1": 8, "drwtNo2": 9, "drwtNo3": 19, "drwtNo4": 25, "drwtNo5": 41, "drwtNo6": 42, "bnusNo": 33}

resources/lotto_history.txt

@@ -1222,3 +1222,4 @@
 1222,4,11,17,22,32,41,34
 1223,16,18,20,32,33,39,26
 1224,9,18,21,27,44,45,28
+1225,8,9,19,25,41,42,33