Add files using upload-large-folder tool
Browse files- scripts/analyze_errors.py +126 -0
- scripts/remove.py +183 -0
scripts/analyze_errors.py
ADDED
|
@@ -0,0 +1,126 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import re
|
| 2 |
+
import os
|
| 3 |
+
from collections import Counter
|
| 4 |
+
|
| 5 |
+
# ================= 配置区域 =================
|
| 6 |
+
# 1. 你的错误日志文件路径
|
| 7 |
+
LOG_FILE = 'copy_errors.log'
|
| 8 |
+
|
| 9 |
+
# 2. 图片路径中 category 和 sequence 的位置模式
|
| 10 |
+
# 假设路径是: data/raw_images/{category}/{sequence}/images/filename.jpg
|
| 11 |
+
# 我们使用正则来提取。
|
| 12 |
+
# 解释:
|
| 13 |
+
# data/raw_images/ -> 匹配前缀 (根据你的实际路径调整,或者用通用的)
|
| 14 |
+
# ([^/]+) -> 第一个捕获组: category (不包含/的任意字符)
|
| 15 |
+
# / -> 分隔符
|
| 16 |
+
# ([^/]+) -> 第二个捕获组: sequence
|
| 17 |
+
# /images/ -> 后面紧跟着 images 目录
|
| 18 |
+
PATH_PATTERN = re.compile(r'data/raw_images/([^/]+)/([^/]+)/images/')
|
| 19 |
+
# ===========================================
|
| 20 |
+
|
| 21 |
+
def analyze_log(log_file):
|
| 22 |
+
if not os.path.exists(log_file):
|
| 23 |
+
print(f"[Error] 找不到日志文件: {log_file}")
|
| 24 |
+
return
|
| 25 |
+
|
| 26 |
+
print(f"[*] 正在分析日志文件: {log_file} ...")
|
| 27 |
+
|
| 28 |
+
failed_categories = set()
|
| 29 |
+
failed_sequences = set()
|
| 30 |
+
|
| 31 |
+
# 用于统计每个 category 和 sequence 缺失了多少张图
|
| 32 |
+
category_counter = Counter()
|
| 33 |
+
sequence_counter = Counter()
|
| 34 |
+
|
| 35 |
+
# 用于存储 {category: [sequence1, sequence2, ...]} 的层级关系
|
| 36 |
+
cat_seq_map = {}
|
| 37 |
+
|
| 38 |
+
with open(log_file, 'r', encoding='utf-8') as f:
|
| 39 |
+
current_path = ""
|
| 40 |
+
|
| 41 |
+
for line in f:
|
| 42 |
+
line = line.strip()
|
| 43 |
+
|
| 44 |
+
# 提取日志中的图片路径行
|
| 45 |
+
# 日志格式: [FAIL] 图片路径: data/raw_images/bowl/...
|
| 46 |
+
if line.startswith("[FAIL] 图片路径:"):
|
| 47 |
+
# 提取路径部分
|
| 48 |
+
path_str = line.split(": ", 1)[1]
|
| 49 |
+
|
| 50 |
+
# 使用正则提取 category 和 sequence
|
| 51 |
+
match = PATH_PATTERN.search(path_str)
|
| 52 |
+
if match:
|
| 53 |
+
category = match.group(1)
|
| 54 |
+
sequence = match.group(2)
|
| 55 |
+
|
| 56 |
+
# 存入集合 (去重)
|
| 57 |
+
failed_categories.add(category)
|
| 58 |
+
failed_sequences.add(sequence)
|
| 59 |
+
|
| 60 |
+
# 计数
|
| 61 |
+
category_counter[category] += 1
|
| 62 |
+
sequence_counter[sequence] += 1
|
| 63 |
+
|
| 64 |
+
# 构建层级关系
|
| 65 |
+
if category not in cat_seq_map:
|
| 66 |
+
cat_seq_map[category] = set()
|
| 67 |
+
cat_seq_map[category].add(sequence)
|
| 68 |
+
else:
|
| 69 |
+
# 如果正则匹配失败,可能路径格式不一样,打印出来看看
|
| 70 |
+
# print(f"[Warning] 无法解析路径结构: {path_str}")
|
| 71 |
+
pass
|
| 72 |
+
|
| 73 |
+
# ================= 输出结果 =================
|
| 74 |
+
print("\n" + "="*50)
|
| 75 |
+
print("分析报告 (Analysis Report)")
|
| 76 |
+
print("="*50)
|
| 77 |
+
|
| 78 |
+
print(f"\n1. 缺失图片涉及的 Category 总数: {len(failed_categories)}")
|
| 79 |
+
print("-" * 30)
|
| 80 |
+
# 按缺失数量从多到少排序
|
| 81 |
+
for cat, count in category_counter.most_common():
|
| 82 |
+
print(f" - {cat}: 缺失 {count} 张图")
|
| 83 |
+
|
| 84 |
+
print(f"\n2. 缺失图片涉及的 Sequence 总数: {len(failed_sequences)}")
|
| 85 |
+
print("-" * 30)
|
| 86 |
+
# 如果 sequence 太多,只打印前 20 个
|
| 87 |
+
top_n = 20
|
| 88 |
+
for seq, count in sequence_counter.most_common(top_n):
|
| 89 |
+
print(f" - {seq}: 缺失 {count} 张图")
|
| 90 |
+
if len(sequence_counter) > top_n:
|
| 91 |
+
print(f" ... (还有 {len(sequence_counter) - top_n} 个 sequence)")
|
| 92 |
+
|
| 93 |
+
print(f"\n3. 详细层级结构 (Category -> Sequence)")
|
| 94 |
+
print("-" * 30)
|
| 95 |
+
for cat in sorted(cat_seq_map.keys()):
|
| 96 |
+
seqs = cat_seq_map[cat]
|
| 97 |
+
print(f"[{cat}] 下有 {len(seqs)} 个有问题的 sequence:")
|
| 98 |
+
# 将 set 转为 list 并排序,方便查看
|
| 99 |
+
sorted_seqs = sorted(list(seqs))
|
| 100 |
+
|
| 101 |
+
# 打印方式:如果很多,就一行打印多个
|
| 102 |
+
# 这里简单处理,每行打印一个,或者你可以改成逗号分隔
|
| 103 |
+
print(f" {', '.join(sorted_seqs)}")
|
| 104 |
+
print("")
|
| 105 |
+
|
| 106 |
+
# 保存结果到文件
|
| 107 |
+
output_file = "missing_stats.txt"
|
| 108 |
+
with open(output_file, 'w', encoding='utf-8') as f_out:
|
| 109 |
+
f_out.write("=== 缺失数据统计 ===\n")
|
| 110 |
+
f_out.write(f"Category 总数: {len(failed_categories)}\n")
|
| 111 |
+
f_out.write(f"Sequence 总数: {len(failed_sequences)}\n\n")
|
| 112 |
+
|
| 113 |
+
f_out.write("=== Category 列表 (格式: 名称 [缺失数量]) ===\n")
|
| 114 |
+
for cat, count in category_counter.most_common():
|
| 115 |
+
f_out.write(f"{cat} [{count}]\n")
|
| 116 |
+
|
| 117 |
+
f_out.write("\n=== Sequence 列表 (按 Category 分组) ===\n")
|
| 118 |
+
for cat in sorted(cat_seq_map.keys()):
|
| 119 |
+
f_out.write(f"\n[{cat}]\n")
|
| 120 |
+
for seq in sorted(list(cat_seq_map[cat])):
|
| 121 |
+
f_out.write(f" - {seq} (缺失 {sequence_counter[seq]} 张)\n")
|
| 122 |
+
|
| 123 |
+
print(f"\n[*] 详细统计已保存至: {output_file}")
|
| 124 |
+
|
| 125 |
+
if __name__ == "__main__":
|
| 126 |
+
analyze_log(LOG_FILE)
|
scripts/remove.py
ADDED
|
@@ -0,0 +1,183 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import json
|
| 3 |
+
import re
|
| 4 |
+
import shutil
|
| 5 |
+
from collections import defaultdict
|
| 6 |
+
|
| 7 |
+
# ===================== 配置区域 =====================
|
| 8 |
+
|
| 9 |
+
# 1. 错误日志文件路径
|
| 10 |
+
LOG_FILE = 'copy_errors.log'
|
| 11 |
+
|
| 12 |
+
# 2. JSONL 文件的根目录 (脚本会去这里找对应的 jsonl 文件进行修改)
|
| 13 |
+
QUESTION_ROOT = 'data/questions'
|
| 14 |
+
|
| 15 |
+
# 3. 过滤条件 (可选)
|
| 16 |
+
# 如果只想删除特定 category 或 sequence 的问题,请在此填入字符串。
|
| 17 |
+
# 如果想处理日志中所有记录的错误,请设置为 None
|
| 18 |
+
TARGET_CATEGORY = None # 例如: 'bowl' 或 None
|
| 19 |
+
TARGET_SEQUENCE = None # 例如: '70_6141_14002' 或 None
|
| 20 |
+
|
| 21 |
+
# ===================================================
|
| 22 |
+
|
| 23 |
+
def parse_log_and_get_ids(log_file, target_cat=None, target_seq=None):
|
| 24 |
+
"""
|
| 25 |
+
解析日志文件,获取需要删除的 {文件名: {ID集合}}
|
| 26 |
+
支持按 category 和 sequence 过滤
|
| 27 |
+
"""
|
| 28 |
+
if not os.path.exists(log_file):
|
| 29 |
+
print(f"[Error] 找不到日志文件: {log_file}")
|
| 30 |
+
return {}
|
| 31 |
+
|
| 32 |
+
print(f"[*] 正在解析日志: {log_file} ...")
|
| 33 |
+
if target_cat:
|
| 34 |
+
print(f" -> 过滤条件 Category: {target_cat}")
|
| 35 |
+
if target_seq:
|
| 36 |
+
print(f" -> 过滤条件 Sequence: {target_seq}")
|
| 37 |
+
|
| 38 |
+
# 存储结构: { 'train_2.jsonl': {'task3_id_1', 'task3_id_2'}, ... }
|
| 39 |
+
files_to_clean = defaultdict(set)
|
| 40 |
+
|
| 41 |
+
# 正则用于提取路径中的 category 和 sequence
|
| 42 |
+
# 假设路径结构: .../raw_images/{category}/{sequence}/images/...
|
| 43 |
+
path_pattern = re.compile(r'data/raw_images/([^/]+)/([^/]+)/images/')
|
| 44 |
+
|
| 45 |
+
current_category = None
|
| 46 |
+
current_sequence = None
|
| 47 |
+
is_target_block = False # 标记当前错误块是否符合过滤条件
|
| 48 |
+
|
| 49 |
+
with open(log_file, 'r', encoding='utf-8') as f:
|
| 50 |
+
for line in f:
|
| 51 |
+
line = line.strip()
|
| 52 |
+
|
| 53 |
+
# 1. 识别错误块的开始 (提取图片路径信息)
|
| 54 |
+
if line.startswith("[FAIL] 图片路径:"):
|
| 55 |
+
path_str = line.split(": ", 1)[1]
|
| 56 |
+
match = path_pattern.search(path_str)
|
| 57 |
+
|
| 58 |
+
if match:
|
| 59 |
+
current_category = match.group(1)
|
| 60 |
+
current_sequence = match.group(2)
|
| 61 |
+
|
| 62 |
+
# 判断是否符合用户过滤条件
|
| 63 |
+
cat_match = (target_cat is None) or (current_category == target_cat)
|
| 64 |
+
seq_match = (target_seq is None) or (current_sequence == target_sequence)
|
| 65 |
+
|
| 66 |
+
is_target_block = cat_match and seq_match
|
| 67 |
+
else:
|
| 68 |
+
# 如果路径解析失败,为了安全起见,默认不处理,或者你可以选择处理
|
| 69 |
+
is_target_block = False
|
| 70 |
+
|
| 71 |
+
# 2. 提取受影响的问题 ID
|
| 72 |
+
# 格式: train_2.jsonl -> ID:task3_209_22099_44906_71
|
| 73 |
+
elif "-> ID:" in line:
|
| 74 |
+
if is_target_block:
|
| 75 |
+
try:
|
| 76 |
+
# 分割文件名和ID
|
| 77 |
+
parts = line.split("-> ID:")
|
| 78 |
+
filename = parts[0].strip()
|
| 79 |
+
q_id = parts[1].strip()
|
| 80 |
+
|
| 81 |
+
files_to_clean[filename].add(q_id)
|
| 82 |
+
except IndexError:
|
| 83 |
+
pass
|
| 84 |
+
|
| 85 |
+
total_ids = sum(len(ids) for ids in files_to_clean.values())
|
| 86 |
+
print(f"[*] 解析完成。共发现 {len(files_to_clean)} 个文件中的 {total_ids} 个问题需要删除。")
|
| 87 |
+
return files_to_clean
|
| 88 |
+
|
| 89 |
+
def clean_jsonl_files(question_root, files_to_clean):
|
| 90 |
+
"""
|
| 91 |
+
遍历目录,找到对应的 jsonl 文件并删除指定 ID 的行
|
| 92 |
+
"""
|
| 93 |
+
if not files_to_clean:
|
| 94 |
+
print("[*] 没有需要删除的内容。")
|
| 95 |
+
return
|
| 96 |
+
|
| 97 |
+
print(f"[*] 开始扫描目录 {question_root} 并执行删除操作...")
|
| 98 |
+
|
| 99 |
+
modified_count = 0
|
| 100 |
+
deleted_lines_count = 0
|
| 101 |
+
|
| 102 |
+
# 遍历所有文件
|
| 103 |
+
for root, dirs, files in os.walk(question_root):
|
| 104 |
+
for file in files:
|
| 105 |
+
# 如果这个文件在我们的清理列表中
|
| 106 |
+
if file in files_to_clean:
|
| 107 |
+
file_path = os.path.join(root, file)
|
| 108 |
+
ids_to_remove = files_to_clean[file]
|
| 109 |
+
|
| 110 |
+
# 执行清理
|
| 111 |
+
removed = process_single_file(file_path, ids_to_remove)
|
| 112 |
+
|
| 113 |
+
if removed > 0:
|
| 114 |
+
modified_count += 1
|
| 115 |
+
deleted_lines_count += removed
|
| 116 |
+
# 从列表中移除已处理的文件(优化后续查找,虽然文件名可能重复,但这里假设文件名唯一对应任务)
|
| 117 |
+
# 注意:如果不同目录下有同名文件 (如 task1/train.jsonl, task2/train.jsonl),
|
| 118 |
+
# 且日志里只记了 "train.jsonl",我们需要对所有叫 train.jsonl 的都检查一遍 ID。
|
| 119 |
+
# 所以这里不从 files_to_clean 中 pop,而是让它继续检查。
|
| 120 |
+
|
| 121 |
+
print("\n" + "="*30)
|
| 122 |
+
print("清理任务完成 Summary:")
|
| 123 |
+
print(f"��改文件数: {modified_count}")
|
| 124 |
+
print(f"删除问题数: {deleted_lines_count}")
|
| 125 |
+
print("="*30)
|
| 126 |
+
|
| 127 |
+
def process_single_file(file_path, ids_to_remove):
|
| 128 |
+
"""
|
| 129 |
+
读取文件,过滤掉在 ids_to_remove 中的行,重写文件
|
| 130 |
+
"""
|
| 131 |
+
temp_file = file_path + '.tmp'
|
| 132 |
+
removed_count = 0
|
| 133 |
+
|
| 134 |
+
try:
|
| 135 |
+
with open(file_path, 'r', encoding='utf-8') as f_in, \
|
| 136 |
+
open(temp_file, 'w', encoding='utf-8') as f_out:
|
| 137 |
+
|
| 138 |
+
for line in f_in:
|
| 139 |
+
line = line.strip()
|
| 140 |
+
if not line:
|
| 141 |
+
continue
|
| 142 |
+
|
| 143 |
+
should_delete = False
|
| 144 |
+
try:
|
| 145 |
+
data = json.loads(line)
|
| 146 |
+
if data.get('id') in ids_to_remove:
|
| 147 |
+
should_delete = True
|
| 148 |
+
except json.JSONDecodeError:
|
| 149 |
+
pass
|
| 150 |
+
|
| 151 |
+
if should_delete:
|
| 152 |
+
removed_count += 1
|
| 153 |
+
else:
|
| 154 |
+
f_out.write(line + '\n')
|
| 155 |
+
|
| 156 |
+
# 如果有删除操作,则用新文件替换旧文件
|
| 157 |
+
if removed_count > 0:
|
| 158 |
+
shutil.move(temp_file, file_path)
|
| 159 |
+
print(f" [Cleaned] {os.path.basename(file_path)}: 删除了 {removed_count} 行")
|
| 160 |
+
else:
|
| 161 |
+
# 如果没变化,删除临时文件
|
| 162 |
+
os.remove(temp_file)
|
| 163 |
+
|
| 164 |
+
except Exception as e:
|
| 165 |
+
print(f"[!] 处理文件 {file_path} 时出错: {e}")
|
| 166 |
+
if os.path.exists(temp_file):
|
| 167 |
+
os.remove(temp_file)
|
| 168 |
+
return 0
|
| 169 |
+
|
| 170 |
+
return removed_count
|
| 171 |
+
|
| 172 |
+
if __name__ == "__main__":
|
| 173 |
+
# 1. 获取删除列表
|
| 174 |
+
clean_map = parse_log_and_get_ids(LOG_FILE, TARGET_CATEGORY, TARGET_SEQUENCE)
|
| 175 |
+
|
| 176 |
+
# 2. 执行删除
|
| 177 |
+
if clean_map:
|
| 178 |
+
# 二次确认防止误删
|
| 179 |
+
confirm = input(f"警告: 即将从原始 jsonl 文件中永久删除数据。\n输入 'yes' 确认执行: ")
|
| 180 |
+
if confirm.lower() == 'yes':
|
| 181 |
+
clean_jsonl_files(QUESTION_ROOT, clean_map)
|
| 182 |
+
else:
|
| 183 |
+
print("[*] 操作已取消。")
|