如何将`pandas`中`DataFrame`的`unstack`方法反转回原始对象？

Question

我想将一个对称相似矩阵 (pd.DataFrame) 转换为带有pd.MultiIndex 的未堆叠pd.Series，然后再转换回pd.DataFrame

这是获取对称pd.DataFrame 的代码以及我尝试反转操作的代码。

我应该使用pivot 吗？我想以另一个名为df_sqr_revert 的pd.DataFrame 结尾，它与原始df_sqr 相同。

有人知道如何扭转这个操作吗？

data = {'sepal_length': {'sepal_length': 1.0, 'sepal_width': 0.44531537502467533, 'petal_length': 0.935877078652436, 'petal_width': 0.9089768166845817}, 'sepal_width': {'sepal_length': 0.44531537502467533, 'sepal_width': 1.0, 'petal_length': 0.2897419517994226, 'petal_width': 0.32172795519309727}, 'petal_length': {'sepal_length': 0.935877078652436, 'sepal_width': 0.2897419517994226, 'petal_length': 1.0, 'petal_width': 0.9813785485254833}, 'petal_width': {'sepal_length': 0.9089768166845817, 'sepal_width': 0.32172795519309727, 'petal_length': 0.9813785485254833, 'petal_width': 1.0}}

df_sqr = pd.DataFrame(data)
#               petal_length  petal_width  sepal_length  sepal_width
# petal_length      1.000000     0.981379      0.935877     0.289742
# petal_width       0.981379     1.000000      0.908977     0.321728
# sepal_length      0.935877     0.908977      1.000000     0.445315
# sepal_width       0.289742     0.321728      0.445315     1.000000
Se_vertical = df_sqr.unstack()
# petal_length  petal_length    1.000000
#               petal_width     0.981379
#               sepal_length    0.935877
#               sepal_width     0.289742
# petal_width   petal_length    0.981379
#               petal_width     1.000000
#               sepal_length    0.908977
#               sepal_width     0.321728
# sepal_length  petal_length    0.935877
#               petal_width     0.908977
#               sepal_length    1.000000
#               sepal_width     0.445315
# sepal_width   petal_length    0.289742
#               petal_width     0.321728
#               sepal_length    0.445315
#               sepal_width     1.000000
# dtype: float64

# df_sqr_revert = Se_vertical.stack()
# AttributeError: 'Series' object has no attribute 'stack'

Answer 1

自相矛盾的是，您在这里需要的是第二次 unstack 调用：

In [14]: df
Out[14]: 
              sepal_length  sepal_width  petal_length  petal_width
petal_length      0.935877     0.289742      1.000000     0.981379
petal_width       0.908977     0.321728      0.981379     1.000000
sepal_length      1.000000     0.445315      0.935877     0.908977
sepal_width       0.445315     1.000000      0.289742     0.321728


In [13]: df_sqr.unstack().unstack()
Out[13]: 
              petal_length  petal_width  sepal_length  sepal_width
sepal_length      0.935877     0.908977      1.000000     0.445315
sepal_width       0.289742     0.321728      0.445315     1.000000
petal_length      1.000000     0.981379      0.935877     0.289742
petal_width       0.981379     1.000000      0.908977     0.321728

documentation 提到在系列等于枢轴的情况下取消堆叠，就像您在问题中怀疑的那样。

---

奖金

只是因为我很好奇，当我们为列和索引标签添加前缀时，stack 和 unstack 之间的区别会变得更加明显：

In [17]: df.columns  = [f'columns_{i}' for i in df.columns]

In [18]: df.index  = [f'index_{i}' for i in df.index]

.stack() 将 row-index 放在多索引的最左边：

In [20]: df.stack()
Out[20]: 
index_petal_length  columns_sepal_length    0.935877
                    columns_sepal_width     0.289742
                    columns_petal_length    1.000000
                    columns_petal_width     0.981379
index_petal_width   columns_sepal_length    0.908977
                    columns_sepal_width     0.321728
                    columns_petal_length    0.981379
                    columns_petal_width     1.000000
index_sepal_length  columns_sepal_length    1.000000
                    columns_sepal_width     0.445315
                    columns_petal_length    0.935877
                    columns_petal_width     0.908977
index_sepal_width   columns_sepal_length    0.445315
                    columns_sepal_width     1.000000
                    columns_petal_length    0.289742
                    columns_petal_width     0.321728
dtype: float64

.unstack()将列索引作为多索引的最左一级：

In [21]: df.unstack()
Out[21]: 
columns_sepal_length  index_petal_length    0.935877
                      index_petal_width     0.908977
                      index_sepal_length    1.000000
                      index_sepal_width     0.445315
columns_sepal_width   index_petal_length    0.289742
                      index_petal_width     0.321728
                      index_sepal_length    0.445315
                      index_sepal_width     1.000000
columns_petal_length  index_petal_length    1.000000
                      index_petal_width     0.981379
                      index_sepal_length    0.935877
                      index_sepal_width     0.289742
columns_petal_width   index_petal_length    0.981379
                      index_petal_width     1.000000
                      index_sepal_length    0.908977
                      index_sepal_width     0.321728
dtype: float64