| import pandas as pd | |
| import os | |
| # 指定数据目录路径 | |
| data_dir = '/home/yiyangai/stephenqs/datasets/physics_big/data' | |
| # 初始化一个空的DataFrame，用来存储筛选出的样本 | |
| combined_df = pd.DataFrame() | |
| # 遍历指定目录中的所有文件 | |
| for i in range(1, 128): # 文件名为00001--00127.parquet | |
| file_name = f'{i:05d}.parquet' | |
| file_path = os.path.join(data_dir, file_name) | |
| # 读取当前的parquet文件 | |
| df = pd.read_parquet(file_path) | |
| # 筛选出images字段不为空的样本 | |
| non_empty_images_df = df[df['images'].apply(lambda x: len(x) > 0)] | |
| # 只取前3个样本 | |
| selected_samples = non_empty_images_df.head(3) | |
| # 将筛选出的样本添加到合并的DataFrame中 | |
| combined_df = pd.concat([combined_df, selected_samples], ignore_index=True) | |
| # 保存合并后的结果到一个新的Parquet文件 | |
| output_file_path = os.path.join(data_dir, 'combined_images_non_empty.parquet') | |
| combined_df.to_parquet(output_file_path) | |
| print(f"已将筛选出的样本保存到 {output_file_path}") | |