Upload tag_char_info.py with huggingface_hub
Browse files- tag_char_info.py +95 -0
tag_char_info.py
ADDED
@@ -0,0 +1,95 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
import json
|
3 |
+
from tqdm import tqdm
|
4 |
+
from concurrent.futures import ProcessPoolExecutor
|
5 |
+
from collections import Counter
|
6 |
+
|
7 |
+
if __name__ == "__main__":
|
8 |
+
# 文件路径
|
9 |
+
danbooru_parquets_path = "/mnt/data/Booru-parquets/danbooru.parquet"
|
10 |
+
danbooru_parquets_path_add = "/mnt/data/danbooru_newest-all/table.parquet"
|
11 |
+
|
12 |
+
# 读取 Parquet 文件
|
13 |
+
df1 = pd.read_parquet(danbooru_parquets_path)
|
14 |
+
df2 = pd.read_parquet(danbooru_parquets_path_add)
|
15 |
+
df = pd.concat([df1, df2], ignore_index=True)
|
16 |
+
print(df.columns)
|
17 |
+
print(df.head())
|
18 |
+
|
19 |
+
# 获取所有 tag_string_character的list
|
20 |
+
tag_string_character = df['tag_string_character'].unique()
|
21 |
+
|
22 |
+
CHARACTER_TAG_LIST = []
|
23 |
+
for tag_list in tqdm(tag_string_character):
|
24 |
+
if tag_list is None:
|
25 |
+
continue # 跳过 None 值
|
26 |
+
for tag in tag_list.split(" "):
|
27 |
+
CHARACTER_TAG_LIST.append(tag)
|
28 |
+
CHARACTER_TAG_LIST = list(set(CHARACTER_TAG_LIST))
|
29 |
+
CHARACTER_TAG_LIST.remove("")
|
30 |
+
print(len(CHARACTER_TAG_LIST))
|
31 |
+
|
32 |
+
def map_function(tag):
|
33 |
+
"""映射函数:获取每个角色的 ID 和相关文本的词频"""
|
34 |
+
# 使用布尔���引来选择包含特定 tag 的行
|
35 |
+
tag_data = df[df['tag_string_character'].str.contains(tag, na=False, regex=False)]
|
36 |
+
|
37 |
+
tag_id = tag_data['id'].unique()
|
38 |
+
word_counts = Counter()
|
39 |
+
|
40 |
+
# 假设有一列包含文本,如 'description',替换为实际列名
|
41 |
+
text_column = 'tag_string_general' # 替换为实际列名
|
42 |
+
for text in tag_data[text_column]:
|
43 |
+
if isinstance(text, str): # 确保文本不是 None
|
44 |
+
words = text.split(" ") # 根据需要进行分词
|
45 |
+
word_counts.update(words)
|
46 |
+
|
47 |
+
return tag, tag_id, word_counts
|
48 |
+
|
49 |
+
def reduce_function(results):
|
50 |
+
"""归约函数:合并所有结果"""
|
51 |
+
tag_string_character_id = {}
|
52 |
+
|
53 |
+
|
54 |
+
for tag, ids, word_counts in results:
|
55 |
+
tag_string_character_id[tag] = {}
|
56 |
+
tag_string_character_id[tag]["ids"] = ids.tolist() # 将 NumPy 数组转换为列表
|
57 |
+
tag_string_character_id[tag]["word_counts"] = dict(word_counts) # 将词频转换为字典
|
58 |
+
|
59 |
+
return tag_string_character_id
|
60 |
+
|
61 |
+
with open("character_tag_list.txt", "w") as f:
|
62 |
+
f.write("\n".join(CHARACTER_TAG_LIST))
|
63 |
+
# CHARACTER_TAG_LIST = CHARACTER_TAG_LIST[:60]
|
64 |
+
# print(CHARACTER_TAG_LIST)
|
65 |
+
# 使用多进程并行处理
|
66 |
+
with ProcessPoolExecutor(max_workers=32) as executor:
|
67 |
+
# 映射阶段
|
68 |
+
mapped_results = list(tqdm(executor.map(map_function, CHARACTER_TAG_LIST), total=len(CHARACTER_TAG_LIST)))
|
69 |
+
|
70 |
+
# 归约阶段
|
71 |
+
tag_string_character_id = reduce_function(mapped_results)
|
72 |
+
|
73 |
+
# 展示前五个艺术家的 ID 和词频
|
74 |
+
# print(list(tag_string_character_id.keys())[:5])
|
75 |
+
# print(list(tag_string_character_id.values())[:5])
|
76 |
+
|
77 |
+
|
78 |
+
# 保存 ID,词频 为 JSON
|
79 |
+
with open("character_id_with_word_counts.json", "w") as f:
|
80 |
+
json.dump(tag_string_character_id, f)
|
81 |
+
|
82 |
+
|
83 |
+
|
84 |
+
# 保存艺术家 ID 和词频为 Parquet
|
85 |
+
character_data = []
|
86 |
+
for tag, data in tag_string_character_id.items():
|
87 |
+
character_data.append({
|
88 |
+
'tag': tag,
|
89 |
+
'count':len(data["ids"]),
|
90 |
+
'id': data["ids"],
|
91 |
+
'word_counts': data["word_counts"]
|
92 |
+
})
|
93 |
+
|
94 |
+
artist_df = pd.DataFrame(character_data)
|
95 |
+
artist_df.to_parquet("char_id_with_word_counts.parquet")
|