Download scripts/_scan.py from Jannchie/danbooru_metadata: direct link, hf CLI and curl.
- Browser
- Download file 23.5 kB
-
https://huggingface.co/datasets/Jannchie/danbooru_metadata/resolve/main/scripts/_scan.py
- Command line
-
hf download hf://datasets/Jannchie/danbooru_metadata/scripts/_scan.py
-
curl -L -o _scan.py https://huggingface.co/datasets/Jannchie/danbooru_metadata/resolve/main/scripts/_scan.py
23.5 kB
| #!/usr/bin/env python3 | |
| r"""生成「Danbooru 各 series(版权作品)累计投稿数」的 bar chart race。 | |
| 一次扫描 → 缓存月度计数 → 同时产出中英两份 CSV(danbooru-series.csv / danbooru-series-zh.csv)。 | |
| 数据源是本地 Danbooru 元数据 SQLite(posts 表,约 1100 万条)。一个 series = Danbooru 的 | |
| copyright 标签(tag_string_copyright,空格分隔)。「投稿数」= 该 series 出现在多少张 post 上, | |
| 按 post 的 created_at 月份累计。 | |
| 同系列合并(关键):Danbooru 把一个大系列拆成很多子标签(fate/grand_order、fate/stay_night、 | |
| fate_(series)…;idolmaster_cinderella_girls…;pokemon_sv…),且有 tag implication,一稿常同时 | |
| 带「子标签 + 系列伞标签」。所以: | |
| · 用前缀规则把子标签归并到规范系列键(见 FRANCHISES,group_of())。 | |
| · 计数按「每稿对每个系列至多计一次」去重——直接把子标签计数相加会把同时带多个子标签的稿件重复计。 | |
| 未列入 FRANCHISES 的标签各自独立成组(组键 = 标签本身)。 | |
| 为什么用累计(cumulative):bar chart race 表达「存量随时间此消彼长」,和仓库里 stocks/gdp/llm 一样 | |
| 是 level-over-time,累计柱单调增长、排名随热度迁移最耐看。 | |
| 同时再出一个「增速榜」(danbooru-series-growth*.csv):value = 滚动 TRAILING_WINDOW 个月新增和 | |
| (默认 12=近一年滚动,最耐看;按月窗太抖故弃用)。能看谁正当红、死掉的系列会自然淡出。两个榜各出 | |
| 中英两版,一次 --from-cache 全产出。 | |
| SFW 过滤:扫描时按 post 的 rating 分桶缓存(g/s/q/e 四级),故一份缓存能同时派生「全部 rating」 | |
| (ALL)与「仅 SFW」(SFW_RATINGS=g+s,对齐 safebooru.donmai.us)两套,无需为 SFW 单独重扫。 | |
| SFW 版文件名插 `-sfw`(如 danbooru-series-sfw.csv、danbooru-series-growth-sfw-zh.csv)。口径取 g+s | |
| 对 2023 年制度迁移鲁棒:老制度 s=safe、新制度新增 g 而 s 改指 sensitive,但 q/e 在两个时代恒为 | |
| NSFW,故「排除 q/e」的桶边界含义一致(若改成仅 g,2023 前的安全内容当时多标 s,会被错杀)。 | |
| 注意:top-N 选取仍按全部 rating 的近似频率(见 select_groups),故某个 SFW 占比极高但总量在 | |
| top-N 之外的系列可能缺席 SFW 榜;调大 --top-n 可放宽。 | |
| CSV 列:series(显示名,中/英两版不同), tag(规范系列键,两版一致,datasets.ts 按它查主题色), | |
| count(值), date(Unix 秒,月 1 号)。 | |
| 口径与取舍: | |
| · 默认排除 `original`(原创/无版权标记,量级碾压)。 | |
| · 只取合并后总量 top-N(默认 60)个系列。选取用频率表近似排序,真实值来自去重扫描。 | |
| · 中文名:FRANCHISES 里写死规范译名;独立标签取 copyright_name_map.json 的 zh_hans→zh_hant, | |
| 缺失回退 ZH_OVERRIDES→英文。 | |
| · 月度采样:某月 value = 截至该月末的累计投稿数,时间戳记为该月 1 号;系列首投前不出现(淡入)。 | |
| 全表扫描 47GB(无覆盖索引),WSL /mnt 上 ~5–10k rows/s。扫一次把月度计数缓存到 | |
| scripts/data/danbooru-series-monthly.json;之后只调显示名/语言用 --from-cache 秒级重生成 | |
| (注意:改 FRANCHISES 分组会影响计数,必须重扫,不能用缓存)。 | |
| 用法: | |
| python3 scripts/update-danbooru-series.py # 全表扫描 + 写缓存 + 中英两份 CSV | |
| python3 scripts/update-danbooru-series.py --from-cache # 跳过扫描,从缓存重生成(仅改显示名时) | |
| python3 scripts/update-danbooru-series.py --limit 500000 # 小样验证管线 | |
| WSL 空间不足、库在 Windows 盘(E:)时:把重扫放宿主机原生跑,只回传几百 KB 缓存—— | |
| # ① WSL 里把脚本拷到 Windows 侧(避免 UNC/CWD 问题) | |
| cp scripts/update-danbooru-series.py /mnt/e/danbooru_metadata/_scan.py | |
| # ② 用 Windows Python 原生扫描(比 WSL 读 /mnt 的 9p 快 3–4×),只导缓存 | |
| python.exe 'E:\danbooru_metadata\_scan.py' --scan-only \ | |
| --database 'E:\danbooru_metadata\data\db\danbooru_metadata.db' \ | |
| --freq-csv 'E:\danbooru_metadata\data\outputs\tags\tag_frequency_copyright.csv' \ | |
| --cache 'E:\danbooru_metadata\series_cache.json' | |
| # ③ 回 WSL:缓存就位 + 出 CSV(小而快) | |
| cp /mnt/e/danbooru_metadata/series_cache.json scripts/data/danbooru-series-monthly.json | |
| python3 scripts/update-danbooru-series.py --from-cache | |
| 依赖:仅标准库。 | |
| """ | |
| from __future__ import annotations | |
| import argparse | |
| import calendar | |
| import csv | |
| import io | |
| import json | |
| import re | |
| import sqlite3 | |
| import sys | |
| import time | |
| from collections import defaultdict | |
| from pathlib import Path | |
| REPO = Path(__file__).resolve().parent.parent | |
| DB_PATH = Path('/mnt/e/danbooru_metadata/data/db/danbooru_metadata.db') | |
| DATA = Path('/mnt/e/danbooru_metadata/data/outputs') | |
| # 增速榜:value = 滚动 TRAILING_WINDOW 个月的新增和。1=当月新增(最跟手);12=近一年滚动(最平滑)。 | |
| # 实践下来按月窗太抖、季节噪声大,按年(12 月)窗最耐看,故定为 12。 | |
| TRAILING_WINDOW = 12 | |
| # 各 tag 维度共用同一套扫描/缓存/出图,只是扫不同列、用不同频率表/译名表/输出名: | |
| # copyright(作品)—— 合并同系列(FRANCHISES),排除原创/伞标签/VTuber 泛类。 | |
| # character(角色)—— 不合并(角色变体量少,主标签已占绝大多数),无排除。 | |
| # original / honkai_(series):见 BLOCKLIST 注释(伞标签与具体作品重复,故排除)。 | |
| FIELDS = { | |
| 'copyright': dict( | |
| column='tag_string_copyright', | |
| freq=DATA / 'tags/tag_frequency_copyright.csv', | |
| name_map=DATA / 'translations/copyright_name_map.json', | |
| out='danbooru-series', merge=True, | |
| blocklist={'original', 'honkai_(series)', 'indie_virtual_youtuber'}, | |
| ), | |
| 'character': dict( | |
| column='tag_string_character', | |
| freq=DATA / 'tags/tag_frequency_character.csv', | |
| name_map=DATA / 'translations/character_name_map.json', | |
| out='danbooru-character', merge=False, | |
| # 同一角色被拆成主标签 + 别名/子形态标签(且 name_map 常给同名)→ 折叠掉次要 tag, | |
| # 只留规范主 tag 一条。不相加(半灵几乎总与妖梦共现、saber 与 artoria 大量共现,相加会重复计)。 | |
| # saber_(fate) ⊂ artoria_pendragon_(fate)(同为阿尔托莉雅,zh 同名) | |
| # konpaku_youmu_(ghost) ⊂ konpaku_youmu(妖梦的半灵,en 同名 Konpaku Youmu) | |
| blocklist={'saber_(fate)', 'konpaku_youmu_(ghost)'}, | |
| ), | |
| } | |
| # 运行时由 --field 填入(group_of / select_groups / scan 读取)。 | |
| _COLUMN = 'tag_string_copyright' | |
| _MERGE = True | |
| _BLOCKLIST: set[str] = set() | |
| PUB_DIRS = ('apps/playground/public', 'apps/studio/public') | |
| # SFW 口径:保留的 rating(Danbooru 四级制 g=general / s=sensitive / q=questionable / e=explicit)。 | |
| # 取 g+s(对齐 safebooru.donmai.us)。对 2023 年的制度迁移鲁棒:老制度 s=safe、新制度新增 g 而 s 改指 | |
| # sensitive,但两个时代里 q/e 始终是 NSFW,故「排除 q/e」这条桶边界在新旧数据上含义一致。改这里即换阈值 | |
| # (如仅 ('g',) 最严格),无需重扫——缓存按 rating 分桶存了全部四级。 | |
| SFW_RATINGS = ('g', 's') | |
| def out_file(base: str, *, growth: bool, sfw: bool, zh: bool) -> list[Path]: | |
| """输出 CSV 路径(playground + studio 两份)。后缀顺序固定 -growth → -sfw → -zh。""" | |
| sfx = ('-growth' if growth else '') + ('-sfw' if sfw else '') + ('-zh' if zh else '') | |
| return [REPO / d / f'{base}{sfx}.csv' for d in PUB_DIRS] | |
| # 同系列合并规则:(规范键, 英文名, 中文名, [标签前缀...])。 | |
| # group_of(tag):命中任一前缀 → 规范键;否则标签独立成组。前缀都挑过、不会误伤其它系列。 | |
| FRANCHISES = [ | |
| ('touhou', 'Touhou Project', '东方Project', ['touhou']), | |
| ('fate', 'Fate', 'Fate', ['fate/', 'fate_(']), | |
| ('idolmaster', 'THE iDOLM@STER', '偶像大师', ['idolmaster', 'gakuen_idolmaster']), | |
| ('love_live', 'Love Live!', 'Love Live!', ['love_live', 'link!_like!_love_live']), | |
| ('pokemon', 'Pokémon', '宝可梦', ['pokemon']), | |
| ('final_fantasy', 'Final Fantasy', '最终幻想', ['final_fantasy']), | |
| ('fire_emblem', 'Fire Emblem', '火焰纹章', ['fire_emblem']), | |
| ('honkai_star_rail', 'Honkai: Star Rail', '崩坏:星穹铁道', ['honkai:_star_rail', 'honkai_star_rail']), | |
| ('honkai_impact_3rd', 'Honkai Impact 3rd', '崩坏3', ['honkai_impact_3rd', 'honkai_impact']), | |
| ('bang_dream', 'BanG Dream!', 'BanG Dream!', ['bang_dream']), | |
| ('gundam', 'Gundam', '高达', ['gundam']), | |
| ('persona', 'Persona', '女神异闻录', ['persona']), | |
| ('umamusume', 'Umamusume', '赛马娘', ['umamusume']), | |
| ('sword_art_online', 'Sword Art Online', '刀剑神域', ['sword_art_online']), | |
| ('madoka', 'Puella Magi Madoka Magica', '魔法少女小圆', ['mahou_shoujo_madoka_magica']), | |
| ('granblue_fantasy', 'Granblue Fantasy', '碧蓝幻想', ['granblue']), | |
| ('danganronpa', 'Danganronpa', '弹丸论破', ['danganronpa']), | |
| ('jojo', "JoJo's Bizarre Adventure", 'JOJO的奇妙冒险', ['jojo_no_kimyou_na_bouken']), | |
| ('precure', 'Pretty Cure', '光之美少女', ['precure']), | |
| ('hololive', 'hololive', 'hololive', ['hololive']), | |
| ('nijisanji', 'Nijisanji', '彩虹社', ['nijisanji']), | |
| ('zelda', 'The Legend of Zelda', '塞尔达传说', ['the_legend_of_zelda']), | |
| ('kantai_collection', 'Kantai Collection', '舰队Collection', ['kantai_collection']), | |
| ('project_moon', 'Project Moon', 'Project Moon', ['project_moon', 'limbus_company', | |
| 'lobotomy_corporation', 'library_of_ruina']), | |
| ('xenoblade', 'Xenoblade Chronicles', '异度神剑', ['xenoblade']), | |
| ('world_witches', 'World Witches', '强袭魔女', ['world_witches', 'strike_witches']), | |
| ('lyrical_nanoha', 'Magical Girl Lyrical Nanoha', '魔法少女奈叶', | |
| ['lyrical_nanoha', 'mahou_shoujo_lyrical_nanoha']), | |
| ('haruhi', 'The Melancholy of Haruhi Suzumiya', '凉宫春日的忧郁', ['suzumiya_haruhi']), | |
| ('naruto', 'Naruto', '火影忍者', ['naruto']), | |
| ] | |
| FRANCHISE_PREFIXES = [(pfx, key) for key, _, _, pfxs in FRANCHISES for pfx in pfxs] | |
| FRANCHISE_EN = {key: en for key, en, _, _ in FRANCHISES} | |
| FRANCHISE_ZH = {key: zh for key, _, zh, _ in FRANCHISES} | |
| # 显示名中文兜底(按 tag 查,作品 / 角色通用,键互不冲突):name_map 缺 zh,或俗称比官方译名通行的。 | |
| ZH_OVERRIDES = { | |
| # 作品 | |
| 'kemono_friends': '兽娘动物园', | |
| 'goddess_of_victory:_nikke': '胜利女神:NIKKE', | |
| 'league_of_legends': '英雄联盟', | |
| 'mario_(series)': '超级马里奥', | |
| 'street_fighter': '街头霸王', | |
| 'dragon_ball': '龙珠', # 大陆官方译名(台译「七龙珠」) | |
| # 角色(character_name_map 缺 zh) | |
| 'komeiji_satori': '古明地觉', | |
| 'doodle_sensei_(blue_archive)': '涂鸦先生', | |
| } | |
| # 独立标签的英文兜底:name_map 的 en 缺失或不够地道时覆盖。 | |
| EN_OVERRIDES = { | |
| 'k-on!': 'K-On!', | |
| 'toaru_majutsu_no_index': 'A Certain Magical Index', | |
| 'mario_(series)': 'Super Mario', | |
| 'sousou_no_frieren': "Frieren: Beyond Journey's End", | |
| 'splatoon_(series)': 'Splatoon', | |
| 'bocchi_the_rock!': 'Bocchi the Rock!', | |
| } | |
| YM_RE = re.compile(r'^\d{4}-\d{2}') | |
| # 去掉显示名末尾的「(系列)」「(anime)」「(系列)」等括号后缀:name_map 派生名常带 Danbooru 标签 | |
| # 的消歧后缀,作品名里不该出现。只清理 name_map / prettify 的兜底名,手写的 FRANCHISES/OVERRIDES 不动。 | |
| PAREN_SUFFIX_RE = re.compile(r'\s*[\((][^))]*[\))]\s*$') | |
| _gcache: dict[str, str] = {} | |
| def strip_paren(name: str) -> str: | |
| cleaned = PAREN_SUFFIX_RE.sub('', name).strip() | |
| return cleaned or name | |
| def group_of(tag: str) -> str: | |
| """标签 → 规范键(copyright 命中 FRANCHISES 前缀则归并;其它维度不合并,标签即键)。带缓存。""" | |
| if not _MERGE: | |
| return tag | |
| g = _gcache.get(tag) | |
| if g is None: | |
| g = tag | |
| for pfx, key in FRANCHISE_PREFIXES: | |
| if tag.startswith(pfx): | |
| g = key | |
| break | |
| _gcache[tag] = g | |
| return g | |
| def parse_args() -> argparse.Namespace: | |
| p = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter) | |
| p.add_argument('--field', choices=tuple(FIELDS), default='copyright', help='扫哪个维度:作品 / 角色') | |
| p.add_argument('--database', type=Path, default=DB_PATH) | |
| p.add_argument('--freq-csv', type=Path, default=None, help='默认按 --field 取对应频率表') | |
| p.add_argument('--name-map', type=Path, default=None, help='默认按 --field 取对应译名表') | |
| p.add_argument('--top-n', type=int, default=60, help='纳入的系列数(合并后,按近似总量)') | |
| p.add_argument('--from-cache', action='store_true', help='跳过扫描,从缓存重生成(仅改显示名时)') | |
| p.add_argument('--scan-only', action='store_true', | |
| help='只扫描+导缓存就退出,不读 name_map / 不写 CSV。用于在宿主机原生跑重扫') | |
| p.add_argument('--cache', type=Path, default=None, help='缓存 JSON 路径(默认 scripts/data/)') | |
| p.add_argument('--batch-size', type=int, default=100_000) | |
| p.add_argument('--limit', type=int, default=None, help='只扫前 N 行(验证用)') | |
| return p.parse_args() | |
| def select_groups(freq_csv: Path, top_n: int) -> list[str]: | |
| """按合并后近似总量(子标签计数求和,仅用于排序)取 top-N 规范系列键。""" | |
| approx: dict[str, int] = defaultdict(int) | |
| with freq_csv.open(encoding='utf-8') as f: | |
| reader = csv.reader(f) | |
| next(reader, None) | |
| for row in reader: | |
| if row and row[0] not in _BLOCKLIST: | |
| approx[group_of(row[0])] += int(row[1]) | |
| ranked = sorted(approx.items(), key=lambda kv: kv[1], reverse=True) | |
| return [k for k, _ in ranked[:top_n]] | |
| def prettify(tag: str) -> str: | |
| return ' '.join(w[:1].upper() + w[1:] if w else w for w in tag.replace('_', ' ').split(' ')) | |
| def name_en(key: str, nm: dict) -> str: | |
| if key in FRANCHISE_EN: | |
| return FRANCHISE_EN[key] | |
| if key in EN_OVERRIDES: | |
| return EN_OVERRIDES[key] | |
| return strip_paren(nm.get(key, {}).get('en') or prettify(key)) | |
| def name_zh(key: str, nm: dict) -> str: | |
| if key in FRANCHISE_ZH: | |
| return FRANCHISE_ZH[key] | |
| if key in ZH_OVERRIDES: | |
| return ZH_OVERRIDES[key] | |
| e = nm.get(key, {}) | |
| return strip_paren(e.get('zh_hans') or e.get('zh_hant') or e.get('en') or prettify(key)) | |
| def scan(db: Path, selected: list[str], batch_size: int, limit: int | None) -> dict: | |
| """单遍扫描 posts,按 (year-month, 规范系列键, rating) 计数,每稿对每个系列至多 +1(去重)。 | |
| cell 存 rating→count 的小字典(rating 缺失记为 '?',计入总量但不算 SFW),一次扫描即可派生 | |
| ALL(全部 rating 求和)与 SFW(仅 SFW_RATINGS)两套,无需为 SFW 单独重扫。""" | |
| keep = set(selected) | |
| monthly: dict[str, dict[str, dict[str, int]]] = \ | |
| defaultdict(lambda: defaultdict(lambda: defaultdict(int))) | |
| conn = sqlite3.connect(f'file:{db}?mode=ro', uri=True) | |
| try: | |
| total = conn.execute('SELECT COUNT(*) FROM posts').fetchone()[0] | |
| if limit is not None: | |
| total = min(total, limit) | |
| cur = conn.execute(f'SELECT created_at, rating, {_COLUMN} FROM posts') | |
| seen, t0 = 0, time.time() | |
| while True: | |
| rows = cur.fetchmany(batch_size) | |
| if not rows: | |
| break | |
| for created_at, rating, tagstr in rows: | |
| if not (created_at and tagstr and YM_RE.match(created_at)): | |
| continue | |
| groups = {g for t in tagstr.split() if (g := group_of(t)) in keep} | |
| if groups: | |
| bucket = monthly[created_at[:7]] | |
| r = rating or '?' | |
| for g in groups: | |
| bucket[g][r] += 1 | |
| seen += len(rows) | |
| rate = seen / (time.time() - t0 or 1) | |
| sys.stderr.write(f'\r scanned {seen:,}/{total:,} ({rate:,.0f} rows/s)') | |
| sys.stderr.flush() | |
| if limit is not None and seen >= limit: | |
| break | |
| sys.stderr.write('\n') | |
| finally: | |
| conn.close() | |
| return monthly | |
| def month_range(months: list[str]) -> list[str]: | |
| lo, hi = min(months), max(months) | |
| y, m = int(lo[:4]), int(lo[5:7]) | |
| hy, hm = int(hi[:4]), int(hi[5:7]) | |
| out = [] | |
| while (y, m) <= (hy, hm): | |
| out.append(f'{y:04d}-{m:02d}') | |
| if m < 12: | |
| m += 1 | |
| else: | |
| y, m = y + 1, 1 | |
| return out | |
| def ym_to_unix(ym: str) -> int: | |
| return calendar.timegm((int(ym[:4]), int(ym[5:7]), 1, 0, 0, 0, 0, 0, 0)) | |
| def cell_count(cell, ratings: tuple[str, ...] | None) -> int: | |
| """单月单系列的投稿数:ratings=None 取全部 rating 之和;否则仅取这些 rating。 | |
| 兼容旧缓存(cell 为 int 总量,无 rating 分级):直接返回,调用方对旧缓存不应传 ratings。""" | |
| if isinstance(cell, dict): | |
| return sum(cell.get(r, 0) for r in ratings) if ratings else sum(cell.values()) | |
| return cell # 旧格式:int 即总量 | |
| def cache_has_ratings(monthly: dict) -> bool: | |
| """新缓存的 cell 是 rating→count 字典;旧缓存是 int 总量(无法派生 SFW)。""" | |
| for keyed in monthly.values(): | |
| for cell in keyed.values(): | |
| return isinstance(cell, dict) | |
| return True | |
| def series_values(deltas: list[int], metric: str, window: int) -> list[int]: | |
| """逐月 value:cumulative=累计和;trailing=滚动 window 月新增和(当前投稿速度)。""" | |
| cum, cums = 0, [] | |
| for d in deltas: | |
| cum += d | |
| cums.append(cum) | |
| if metric == 'cumulative': | |
| return cums | |
| return [cums[i] - (cums[i - window] if i - window >= 0 else 0) for i in range(len(cums))] | |
| def usable_months(monthly: dict) -> list[str]: | |
| """补齐空月,并剔除末尾「数据不完整」的月份。 | |
| 抓取常发生在月中,当月投稿数会骤降(如 2026-06 只有上月的 ~12%),月度增速榜里会让最后一帧 | |
| 所有柱断崖式塌方、严重误导。判据:末月总量 < 前三个完整月中位数的一半 → 视为不完整,剔除。""" | |
| months = month_range(list(monthly.keys())) | |
| # 末月完整性判据按总量(与 rating 无关,是抓取时机问题),ALL/SFW 共用同一月轴便于对比。 | |
| vol = {ym: sum(cell_count(c, None) for c in monthly.get(ym, {}).values()) for ym in months} | |
| while len(months) > 4: | |
| ref = sorted(vol[m] for m in months[-4:-1])[1] # 前三个月的中位数 | |
| if ref > 0 and vol[months[-1]] < 0.5 * ref: | |
| months = months[:-1] | |
| else: | |
| break | |
| return months | |
| def build_rows(monthly: dict, selected: list[str], metric: str, name_fn, nm: dict, | |
| ratings: tuple[str, ...] | None = None, window: int = TRAILING_WINDOW) -> list[dict]: | |
| months = usable_months(monthly) | |
| rows: list[dict] = [] | |
| for key in selected: | |
| display = name_fn(key, nm) | |
| deltas = [cell_count(monthly.get(ym, {}).get(key, {}), ratings) for ym in months] | |
| values = series_values(deltas, metric, window) | |
| # 只发「活跃区间」[首个>0, 末个>0]:累计榜末尾恒>0 即到末月;增速榜里死掉的系列自然淡出。 | |
| nz = [i for i, v in enumerate(values) if v > 0] | |
| if not nz: | |
| continue | |
| for i in range(nz[0], nz[-1] + 1): | |
| rows.append({'series': display, 'tag': key, 'count': values[i], | |
| 'date': ym_to_unix(months[i])}) | |
| rows.sort(key=lambda r: (r['date'], r['series'])) | |
| return rows | |
| def write_rows(rows: list[dict], outputs: list[Path]) -> None: | |
| buf = io.StringIO() | |
| w = csv.DictWriter(buf, fieldnames=['series', 'tag', 'count', 'date']) | |
| w.writeheader() | |
| w.writerows(rows) | |
| data = buf.getvalue() | |
| for path in outputs: | |
| path.parent.mkdir(parents=True, exist_ok=True) | |
| path.write_text(data, encoding='utf-8') | |
| print(f' wrote {len(rows):,} rows -> {path}') | |
| def main() -> None: | |
| global _COLUMN, _MERGE, _BLOCKLIST | |
| args = parse_args() | |
| fld = FIELDS[args.field] | |
| _COLUMN, _MERGE, _BLOCKLIST = fld['column'], fld['merge'], fld['blocklist'] | |
| freq_csv = args.freq_csv or fld['freq'] | |
| name_map = args.name_map or fld['name_map'] | |
| cache_path = args.cache or REPO / f"scripts/data/{fld['out']}-monthly.json" | |
| base = fld['out'] | |
| if args.from_cache: | |
| if not cache_path.exists(): | |
| sys.exit(f'cache not found: {cache_path} —— 先不带 --from-cache 跑一次全量扫描') | |
| cached = json.loads(cache_path.read_text(encoding='utf-8')) | |
| # 缓存里可能含后来加入 blocklist 的键;从 selected 过滤掉即可,无需重扫。 | |
| selected = [k for k in cached['selected'] if k not in _BLOCKLIST] | |
| monthly = cached['monthly'] | |
| print(f'[{args.field}] loaded cache: {len(selected)} groups, {len(monthly)} months') | |
| else: | |
| for path in (args.database, freq_csv): | |
| if not path.exists(): | |
| sys.exit(f'not found: {path}') | |
| selected = select_groups(freq_csv, args.top_n) | |
| print(f'[{args.field}] selected {len(selected)} groups (top-{args.top_n})') | |
| monthly = scan(args.database, selected, args.batch_size, args.limit) | |
| print(f'scanned {len(monthly)} months: {min(monthly)} .. {max(monthly)}') | |
| if args.limit is None: | |
| cache_path.parent.mkdir(parents=True, exist_ok=True) | |
| cache_path.write_text(json.dumps({'selected': selected, 'monthly': monthly}), encoding='utf-8') | |
| print(f' cached -> {cache_path}') | |
| # 宿主机只负责「扫描→导缓存」这一重活;显示名映射 / 写 CSV 留给 WSL 侧(小而快)。 | |
| if args.scan_only: | |
| print('scan-only: cache written, skipping name_map / CSV') | |
| return | |
| nm = json.loads(name_map.read_text(encoding='utf-8')) if name_map.exists() else {} | |
| # 一份缓存出全部变体:{累计, 增速} × {ALL 全部 rating, SFW 仅 g+s} × {英, 中}。 | |
| # SFW 需 rating 分级缓存;旧缓存(int cell)只出 ALL,并提示重扫。 | |
| safeties = [('all', None)] | |
| if cache_has_ratings(monthly): | |
| safeties.append(('sfw', SFW_RATINGS)) | |
| else: | |
| print(' 缓存为旧格式(无 rating 分级),跳过 SFW;出 SFW 版需重扫一次') | |
| for metric, growth in (('cumulative', False), ('trailing', True)): | |
| for safety, ratings in safeties: | |
| for zh, name_fn in ((False, name_en), (True, name_zh)): | |
| rows = build_rows(monthly, selected, metric, name_fn, nm, ratings) | |
| tag = f'{metric}/{safety}/{"中" if zh else "en"}' | |
| print(f'[{tag}]:') | |
| write_rows(rows, out_file(base, growth=growth, sfw=safety == 'sfw', zh=zh)) | |
| if __name__ == '__main__': | |
| main() | |