我相信您不想更改输出数据中的dtypes，因此可能的解决方案是分别用DataFrame.dropna和DataFrame.pivot处理每一列，然后通过concat合并在一起：

df['date_value'] = pd.to_datetime(df['date_value'])

df1 = df.dropna(subset=['text_value']).pivot('key','attribute','text_value')
df2 = df.dropna(subset=['numeric_value']).pivot('key','numeric_value')
df3 = df.dropna(subset=['date_value']).pivot('key','date_value')

df = pd.concat([df1,df2,df3],axis=1).reindex(df['attribute'].unique(),axis=1)
print (df)
attribute      order  size  weight  price   segment
key                                                
1         2019-10-02  43.0    22.0   33.0   product
2         2019-11-02  34.0    32.0   89.0  customer

print (df.dtypes)
order      datetime64[ns]
size              float64
weight            float64
price             float64
segment            object
dtype: object

旧答案-所有值都强制转换为字符串：

df['date_value'] = pd.to_datetime(df['date_value'])

df['text_value'] = df['text_value'].fillna(df['numeric_value']).fillna(df['date_value'])

df = df.pivot('key','text_value')
print (df)
attribute                order price   segment size weight
key                                                       
1          1569974400000000000    33   product   43     22
2          1572652800000000000    89  customer   34     32

print (df.dtypes)
order      object
price      object
segment    object
size       object
weight     object
dtype: object

这是解决方案，我想通了

attr_dict = {'order':'date_value','size':'numeric_value','weight':'numeric_value','price':'numeric_value','segment':'text_value'}

output_table = pd.DataFrame()
for attr in attr_dict.keys():
    temp = input_table[input_table['attribute'] == attr][['key',attr_dict[attr]]]
    temp.rename(columns={attr_dict[attr]:attr},inplace=True)
    output_table[attr] = list(temp.values[:,1])
output_table