#!/usr/bin/env python3 """ pattern_mine.py — 写作模式挖掘器 MAP-Elites 质量-多样性搜索，结果写入 evolution/patterns/。 """ import sys, os, re, json, random, hashlib from pathlib import Path from collections import defaultdict from typing import Any sys.path.insert(0, os.path.dirname(os.path.abspath(__file__))) from utils import ( find_project_root, get_evolution_dir, load_evolution, save_evolution, parse_chapters, log_evolution_event, ) class WritingPattern: def __init__(self, name: str, category: str, template: str, complexity: float = 0.5, diversity: float = 0.5, quality_score: float = 0.0): self.name = name self.category = category self.template = template self.complexity = complexity self.diversity = diversity self.quality_score = quality_score self.usage_count = 0 self.success_count = 0 self.fingerprint = hashlib.md5(f"{category}:{template[:100]}".encode()).hexdigest()[:8] def to_dict(self) -> dict: return {"name": self.name, "category": self.category, "template": self.template, "complexity": self.complexity, "diversity": self.diversity, "quality_score": self.quality_score, "usage_count": self.usage_count, "success_count": self.success_count, "fingerprint": self.fingerprint} @classmethod def from_dict(cls, d: dict) -> "WritingPattern": p = cls(d["name"], d["category"], d["template"], d.get("complexity", 0.5), d.get("diversity", 0.5), d.get("quality_score", 0.0)) p.usage_count = d.get("usage_count", 0) p.success_count = d.get("success_count", 0) p.fingerprint = d.get("fingerprint", p.fingerprint) return p class PatternExtractor: HOOK_PATTERNS = [ (r"就在这时[,，](.+?)(?:[。！?]|\n)", "危机降临"), (r"他[忽然突然].*?发现(.+?)(?:[。！?]|\n)", "信息揭示"), (r"(?:但是|然而|可是)(.+?)(?:[。！?]|\n)", "反转"), (r'[\"\'""「].+?[\"\'""」]\s*$', "对话悬念"), (r"(?:明天|明日|天亮|第二).*?(?:就会|将要|会)(.+?)", "时间压力"), (r"(?:他|她).*?(?:决定|选择|下决心)(.+?)", "决定时刻"), (r"(?:远处|门外|身后).*?(?:传来|响起)(.+?)", "威胁到来"), ] PAYOFF_PATTERNS = [ (r"(?:众人|所有人|在场).*?(?:震惊|愕然|呆住)", "打脸爽"), (r"(?:突破|晋升|觉醒|进化|升级).*?(?:成功|完成)", "升级爽"), (r"(?:获得|得到|继承|传承|吸收).+?(?:力量|能力|宝物)", "获得爽"), (r"(?:守护|保护|挡住|拦下).+?(?:身后|背后)", "守护爽"), ] @classmethod def extract_all(cls, text: str, chapter_num: int, score: float = 0) -> list[WritingPattern]: patterns = [] text_clean = text.replace("\n", " ") last_500 = text_clean[-500:] for pattern_re, name in cls.HOOK_PATTERNS: matches = re.findall(pattern_re, last_500) if matches: p = WritingPattern( name=f"钩子_{name}", category="hook", template=matches[0][:200], complexity=min(1.0, len(matches[0]) / 80.0), diversity=random.uniform(0.3, 0.9), quality_score=score, ) p.usage_count = 1 p.success_count = 1 if score >= 75 else 0 patterns.append(p) for pattern_re, name in cls.PAYOFF_PATTERNS: matches = re.findall(pattern_re, text_clean) if matches: p = WritingPattern( name=f"爽点_{name}", category="payoff", template=matches[0][:200], complexity=min(1.0, len(matches) / 5.0), diversity=0.3 + (hash(name) % 70) / 100.0, quality_score=score, ) p.usage_count = len(matches) p.success_count = len(matches) if score >= 75 else 0 patterns.append(p) return patterns class MAPElitesGrid: def __init__(self, bins: int = 5): self.bins = bins self.grid: dict[tuple[int, int], list[WritingPattern]] = defaultdict(list) def _bin_coord(self, complexity: float, diversity: float) -> tuple[int, int]: c = min(self.bins - 1, int(complexity * self.bins)) d = min(self.bins - 1, int(diversity * self.bins)) return (c, d) def insert(self, pattern: WritingPattern) -> bool: coord = self._bin_coord(pattern.complexity, pattern.diversity) for existing in self.grid[coord]: if existing.fingerprint == pattern.fingerprint: existing.usage_count += pattern.usage_count existing.success_count += pattern.success_count if pattern.quality_score > existing.quality_score: existing.quality_score = pattern.quality_score existing.template = pattern.template return False self.grid[coord].append(pattern) return True def get_elite(self, n: int = 3) -> list[WritingPattern]: elites = [] for cell in self.grid.values(): if cell: cell.sort(key=lambda p: p.quality_score, reverse=True) elites.append(cell[0]) elites.sort(key=lambda p: p.quality_score, reverse=True) return elites[:n] def cell_count(self) -> int: return sum(1 for cell in self.grid.values() if cell) # ── 主流程 ───────────────────────────────────────────── def mine_project(project: Path, min_score: int = 70, top_k: int = 20) -> dict: print(f"⛏ 挖掘项目: {project.name}") chapters = parse_chapters(project) if not chapters: print(" ⚠ 未找到章节文件") return {"status": "no_chapters"} grid = MAPElitesGrid(bins=5) total_patterns = 0 high_quality = 0 # 加载项目数据库获取质量评分 from utils import load_project_db db = load_project_db(project) quality_scores = db.get("meta", {}).get("quality_scores", {}) for ch_num, ch_path in sorted(chapters.items()): try: text = ch_path.read_text(encoding="utf-8") except Exception: continue score = quality_scores.get(str(ch_num), 75) patterns = PatternExtractor.extract_all(text, ch_num, score) for p in patterns: total_patterns += 1 if grid.insert(p) and p.quality_score >= min_score: high_quality += 1 elites = grid.get_elite(top_k) # 合并到 evolution/patterns/pattern_library.json existing = load_evolution("patterns/pattern_library.json") or {"patterns": [], "grid_cells": 0} all_patterns = {p.get("fingerprint", ""): p for p in existing.get("patterns", [])} for p in elites: fp = p.fingerprint if fp not in all_patterns or p.quality_score > all_patterns[fp].get("quality_score", 0): all_patterns[fp] = p.to_dict() pattern_list = sorted(all_patterns.values(), key=lambda x: x.get("quality_score", 0), reverse=True) # 保留前 50 个 pattern_list = pattern_list[:50] output = { "patterns": pattern_list, "grid_cells": grid.cell_count(), "last_mined_from": project.name, "total_chapters": len(chapters), } save_evolution("patterns/pattern_library.json", output) print(f" ✓ 分析了 {len(chapters)} 章") print(f" ✓ 发现 {total_patterns} 个模式（高质量 {high_quality}）") print(f" ✓ MAP-Elites 网格: {grid.cell_count()}/25 格") print(f" ✓ 模式库保有 {len(pattern_list)} 个模式") print(f" ✓ 已写入 evolution/patterns/pattern_library.json") return {"status": "ok", "chapters": len(chapters), "patterns_found": total_patterns, "library_size": len(pattern_list)} def show_patterns(category: str | None = None, top_n: int = 10) -> None: data = load_evolution("patterns/pattern_library.json") patterns = data.get("patterns", []) if category: patterns = [p for p in patterns if p.get("category") == category] patterns.sort(key=lambda p: p.get("quality_score", 0), reverse=True) print(f"\n═══ 写作模式库 (evolution/patterns/) ═══") print(f"总计: {len(patterns)} 个模式\n") for i, p in enumerate(patterns[:top_n]): print(f"{'★' if p.get('quality_score', 0) >= 80 else '●'} " f"[{p.get('category', '?')}] {p.get('name', '?')}") print(f" 质量: {p.get('quality_score', 0):.0f} | " f"complexity={p.get('complexity', 0):.2f} diversity={p.get('diversity', 0):.2f}") preview = p.get("template", "")[:80].replace("\n", " ") print(f" {preview}...") print() if __name__ == "__main__": import argparse parser = argparse.ArgumentParser(description="写作模式挖掘器") parser.add_argument("action", choices=["mine", "show"], help="mine=挖掘 | show=展示") parser.add_argument("--project", "-p", default=".") parser.add_argument("--min-score", type=int, default=70) parser.add_argument("--top", type=int, default=10) parser.add_argument("--category", "-c", choices=["hook", "payoff", "dialogue", "pacing", "description", "emotion"]) args = parser.parse_args() if args.action == "mine": root = find_project_root(Path(args.project).resolve()) if not root: print("错误: 未找到小说项目根目录") sys.exit(1) result = mine_project(root, min_score=args.min_score, top_k=args.top) sys.exit(0 if result["status"] == "ok" else 1) elif args.action == "show": show_patterns(category=args.category, top_n=args.top)