python - 使用 apply 将从一列(具有 json 类型)提取的值插入到另一列

Question

我有这个数据集:

userid   sub_id    event
1        NaN       {'score':25, 'sub_id':5}
1        5         {'score':1}

当 sub_id 列为 NaN 时，我想使用以下代码从 event 列中提取此信息:

df['sub_id'] = df.apply(lambda row: 
                        row['event'].split('sub_id')[1] 
                        if pd.isnull(row['sub_id']) 
                        else row['sub_id']) 

但是，我收到此错误:KeyError: ('sub_id', u'发生在索引索引')

我正在尝试获取此数据框:

userid   sub_id    event
1        5         {'score':25, 'sub_id':5}
1        5         {'score':1}

对错误有什么想法，或者对不同解决方案有什么建议吗？

更新

我需要提取嵌套字典元素中的值:

event
{u'POST': {u'{"options_selected":{"Ideas":"0"},"criterion_feedback":{},"overall_feedback":"Feedback_text_goes_here_1"}': [u'']}, u'GET': {}}

我正在使用此代码:

df['POST'] = df['event'].apply(pd.Series)['POST']

创建以下列:

POST
{u'{"options_selected":{"Ideas":"0"},"criterion_feedback":{},"overall_feedback":"Feedback_text_goes_here_1"}': [u'']}

但是，我需要获取 overall_feedback 值。由于 POST 字段的格式问题，以下代码不起作用:

df['POST'].apply(pd.Series)['overall_feedback']

它抛出此错误KeyError:'overall_feedback'

有什么想法吗？

Answer 1

您可以使用combine_first或fillna :

print (type(df.loc[0, 'event']))
<class 'dict'>

df['sub_id'] = df['sub_id'].combine_first(df.event.apply(lambda x: x['score']))
#df['sub_id'] = df['sub_id'].fillna(df.event.apply(lambda x: x['score']))
print (df)
                       event  sub_id  userid
0  {'sub_id': 5, 'score': 5}     5.0       1
1               {'score': 1}     5.0       1

编辑:如果嵌套字典，更快的解决方案是使用双DataFame构造函数和较慢的解决方案双apply与Series:

df = pd.DataFrame({'userid':[1,1],
                   'sub_id':[np.nan, 5],
                   'event':[{'post':{'score':25, 'sub_id':5}},{'post':{'score':1}} ]})

print (df)
                                  event  sub_id  userid
0  {'post': {'sub_id': 5, 'score': 25}}     NaN       1
1                {'post': {'score': 1}}     5.0       1

s = pd.DataFrame(pd.DataFrame(df['event'].values.tolist())['post'].values.tolist())['score']
print (s)
0    25
1     1
Name: score, dtype: int64

s = df['event'].apply(pd.Series)['post'].apply(pd.Series)['score']
print (s)
0    25.0
1     1.0
Name: score, dtype: float64

---

df['sub_id'] = df['sub_id'].combine_first(s)
print (df)
                                  event  sub_id  userid
0  {'post': {'sub_id': 5, 'score': 25}}    25.0       1
1                {'post': {'score': 1}}     5.0       1

编辑1:

要转换为字典，可以使用:

import ast, yaml

df = pd.DataFrame({'userid':[1,1],
                   'sub_id':[np.nan, 5],
                   'event':[{'post':{'score':25, 'sub_id':5}},{'post':{'score':1}} ]})

df.event = df.event.astype(str)
print (type(df.loc[0, 'event']))
<class 'str'>

df['event'] = df['event'].apply(ast.literal_eval)
#df['event'] = df['event'].apply(yaml.load)
print (df)
                                  event  sub_id  userid
0  {'post': {'sub_id': 5, 'score': 25}}     NaN       1
1                {'post': {'score': 1}}     5.0       1

print (type(df.loc[0, 'event']))
<class 'dict'>

编辑2:

d = {u'{"options_selected":{"Ideas":"0"},"criterion_feedback":{},"overall_feedback":"Feedback_text_goes_here_1"}': [u'']}
d1 = {u'{"options_selected":{"Ideas":"2"},"criterion_feedback":{},"overall_feedback":"Feedback_text_goes_here_2"}': [u'']}

df = pd.DataFrame({'userid':[1,1],
                   'sub_id':[np.nan, 5],
                   'event':[d,d1]})

df['event'] = df['event'].astype(str).apply(yaml.load).apply(lambda x : yaml.load(list(x.keys())[0]))

print (type(df.event.iloc[0]))
<class 'dict'>

print (df.event.apply(pd.Series)['overall_feedback'])
0    Feedback_text_goes_here_1
1    Feedback_text_goes_here_2
Name: overall_feedback, dtype: object