从熊猫数据框中提取字典值

Question

我需要从我从 .json 文件导入的数据集中添加一个特征。

这就是它的样子：

f1 = pd.read_json('https://raw.githubusercontent.com/ansymo/msr2013-bug_dataset/master/data/v02/eclipse/short_desc.json')

print(f1.head())


                                               short_desc
1       [{'when': 1002742486, 'what': 'Usability issue...
10      [{'when': 1002742495, 'what': 'API - VCM event...
100     [{'when': 1002742586, 'what': 'Would like a wa...
10000   [{'when': 1014113227, 'what': 'getter/setter c...
100001  [{'when': 1118743999, 'what': 'Create Help Ind...

本质上，我需要将 'short_desc' 作为列名，并使用其正下方的字符串值填充它：'Usability issue...

到目前为止，我已经尝试了以下方法：

f1['desc'] = pd.DataFrame([x for x in f1['short_desc']])

Wrong number of items passed 19, placement implies 1

有没有不使用循环的简单方法来完成此任务？有人能指出这个新手正确的方向吗？

Answer 1

不要初始化数据框并尝试将其分配给列 - 列应为 pd.Series。

您应该直接分配列表推导，如下所示：

f1['desc'] = [x[0]['what'] for x in f1['short_desc']]

---

作为替代方案，我会提出一个不涉及任何 lambda 函数的解决方案，使用 operator 和 pd.Series.apply：

import operator

f1['desc'] = f1.short_desc.apply(operator.itemgetter(0))\
                             .apply(operator.itemgetter('what'))
print(f1.desc.head())

1           Usability issue with external editors (1GE6IRL)
10                   API - VCM event notification (1G8G6RR)
100       Would like a way to take a write lock on a tea...
10000     getter/setter code generation drops "F" in ".....
100001    Create Help Index Fails with seemingly incorre...
Name: desc, dtype: object

Answer 2

或者你可以试试apply（PS：apply考虑时间成本函数）

f1['short_desc'].apply(pd.Series)[0].apply(pd.Series)

Out[864]: 
                                                     what        when   who
1         Usability issue with external editors (1GE6IRL)  1002742486    21
10                 API - VCM event notification (1G8G6RR)  1002742495    10
100     Would like a way to take a write lock on a tea...  1002742586    24
10000   getter/setter code generation drops "F" in ".....  1014113227   331
100001  Create Help Index Fails with seemingly incorre...  1118743999  9571