将非结构化数据解析为panda数据帧

Question

在通过read_csv:导入*.txt文件后，我目前在熊猫数据帧中有以下数据结构

label   text
0   ###24293578 NaN
1   INTRO   Some text...
2   METHODS Some text...
3   METHODS Some text...
4   METHODS Some text...
5   RESULTS Some text...
6   ###24854809 NaN
7   BACKGROUND  Some text...
8   INTRO   Some text...
9   METHODS Some text...
10  METHODS Some text...
11  RESULTS Some text...
12  ###25165090 NaN
13  BACKGROUND  Some text...
14  METHODS Some text...
...

我喜欢实现的是为每一行创建一个运行索引，从标有"###"的id中检索：

id        label       text
24293578  INTRO       Some text...
24293578  METHODS     Some text...
24293578  ...         ...
24854809  BACKGROUND  Some text...
24854809  ...         ...
25165090  BACKGROUND  Some text...
25165090  ...         ...

我目前使用以下代码来转换数据：

m = df['label'].str.contains("###", na=False) 
df['new'] = df['label'].where(m).ffill()
df = df[df['label'] != df['new']].copy()
df['label'] = df.pop('new').str.lstrip('#') + ' ' + df['label']
df[['id','area']] = df['label'].str.split(' ',expand=True)
df = df.drop(columns=['label'])
df

输出：

text            id          area
1   Some text...    24293578    OBJECTIVE
...
6   Some text...    24854809    BACKGROUND
...

它做得很好但我觉得这不是最好的方法有没有一种方法可以让代码写得更干净，或者更高效我也很好奇，a函数是否可以直接嵌入到read_csv步骤中。

谢谢！

Answer 1

在这里，您可以分三步完成：

# put in the label column into id where text is null, and strip out the #. 
# The rest will be NaN
df['id'] = df.loc[df['text'].isnull(),'label'].str.strip('#')
# forward fill in ID
df['id'].ffill(inplace=True)
# Remove the columns where text is null
df.dropna(subset=['text'], inplace=True)
>>> df
label          text        id
1        INTRO  Some text...  24293578
2      METHODS  Some text...  24293578
3      METHODS  Some text...  24293578
4      METHODS  Some text...  24293578
5      RESULTS  Some text...  24293578
7   BACKGROUND  Some text...  24854809
8        INTRO  Some text...  24854809
9      METHODS  Some text...  24854809
10     METHODS  Some text...  24854809
11     RESULTS  Some text...  24854809
13  BACKGROUND  Some text...  25165090
14     METHODS  Some text...  25165090