旋转数据帧后检索普通数据帧

Question

我有一个熊猫数据框架，这是一个旋转的结果。它有多个指数。我想从这个旋转的数据框中得到一个“正常”的数据帧…这样我就可以在新的数据框上做一些正常的操作。在

下面是一个示例：我的数据透视框如下所示：

                      feature_value
feature_type          f1  f2  f3  f4  f5
time         name
2016-05-10   Clay     0   1   30   4  40
2016-05-10   John     0   4   10   4  66
2016-05-10   Mary     0   1   40   4  46
2016-05-10   Boby     2   0   30   4  59
2016-05-10   Lucy     5   8   20   4  41

以下是我想要的新df：

^{pr2}$

我该怎么做？在

旋转的_数据到字典（）如下所示：

{('feature_value', 'f1'): {(Timestamp('2016-05-10'), 'Clay'): 0, (Timestamp('2016-05-10'), 'John'): 0, (Timestamp('2016-05-10'), 'Mary'): 0, (Timestamp('2016-05-10'), 'Boby'): 2, (Timestamp('2016-05-10'), 'Lucy'): 5}, ('feature_value', 'f2'): {(Timestamp('2016-05-10'), 'Clay'): 1, (Timestamp('2016-05-10'), 'John'): 4, (Timestamp('2016-05-10'), 'Mary'): 1, (Timestamp('2016-05-10'), 'Boby'): 0, (Timestamp('2016-05-10'), 'Lucy'): 8}, ('feature_value', 'f3'): {(Timestamp('2016-05-10'), 'Clay'): 30, (Timestamp('2016-05-10'), 'John'): 10, (Timestamp('2016-05-10'), 'Mary'): 40, (Timestamp('2016-05-10'), 'Boby'): 30, (Timestamp('2016-05-10'), 'Lucy'): 20}, ('feature_value', 'f4'): {(Timestamp('2016-05-10'), 'Clay'): 4, (Timestamp('2016-05-10'), 'John'): 4, (Timestamp('2016-05-10'), 'Mary'): 4, (Timestamp('2016-05-10'), 'Boby'): 4, (Timestamp('2016-05-10'), 'Lucy'): 4}, ('feature_value', 'f5'): {(Timestamp('2016-05-10'), 'Clay'): 40, (Timestamp('2016-05-10'), 'John'): 66, (Timestamp('2016-05-10'), 'Mary'): 46, (Timestamp('2016-05-10'), 'Boby'): 59, (Timestamp('2016-05-10'), 'Lucy'): 41}}

Answer 1

暴力：

df.columns = df.columns.droplevel()
df = df.reset_index()

Answer 2

调用pivot_table时，请确保指定了values参数：

df.pivot_table(index=['time', 'name'], columns=['feature_type'], 
               values='feature_value')

如果没有values='feature_value'，您将得到一个多索引列索引（可能的话）只有一个外部级别，例如'feature_value'。在

df.pivot_table(index=['time', 'name'], ...)还将返回一个具有time和name级别的多索引行索引的数据帧。要使这些索引级别成为常规列，请调用reset_index()：

^{pr2}$

---

例如

import numpy as np
import pandas as pd
np.random.seed(2016)

N = 10
df = pd.DataFrame(
    {'time': np.random.choice(pd.date_range('2016-05-10', '2016-05-12'), size=N),
     'name': np.random.choice(['Clay', 'John', 'Mary', 'Boby', 'Lucy'], size=N),
     'feature_type': np.random.choice(['f{}'.format(i) for i in range(1,6)], size=N),
     'feature_value': np.random.randint(100, size=N)})

orig = df.pivot_table(index=['time', 'name'], columns=['feature_type'])
print(orig)

alt = df.pivot_table(index=['time', 'name'], 
                     columns=['feature_type'],
                     values='feature_value').reset_index()
alt.columns.name = None
print(alt)

orig如下所示：

                feature_value                        
feature_type               f1    f2    f3    f4    f5
time       name                                      
2016-05-10 John           NaN  50.0   NaN   NaN  91.0
           Lucy           NaN   NaN   NaN  28.0   NaN
           Mary           NaN   NaN  19.0   NaN  27.0
2016-05-11 Clay           2.0   NaN   NaN   NaN   NaN
           Lucy          24.0   NaN   NaN   NaN   NaN
2016-05-12 Boby           NaN  16.0   NaN   NaN   NaN
           John           NaN   NaN   NaN   NaN  62.0
           Mary           NaN   NaN   NaN  84.0   NaN

而alt看起来像

        time  name    f1    f2    f3    f4    f5
0 2016-05-10  John   NaN  50.0   NaN   NaN  91.0
1 2016-05-10  Lucy   NaN   NaN   NaN  28.0   NaN
2 2016-05-10  Mary   NaN   NaN  19.0   NaN  27.0
3 2016-05-11  Clay   2.0   NaN   NaN   NaN   NaN
4 2016-05-11  Lucy  24.0   NaN   NaN   NaN   NaN
5 2016-05-12  Boby   NaN  16.0   NaN   NaN   NaN
6 2016-05-12  John   NaN   NaN   NaN   NaN  62.0
7 2016-05-12  Mary   NaN   NaN   NaN  84.0   NaN