我如何真正使用pandas DataFrame的`ix`方法？

Question

注意:因为我问过这个问题,.ix但仍然存在但是或多或少已被.loc基于标签的索引和.iloc位置索引所取代.

---

在阅读了文档之后ix的DataFrames方法,我对我的MultiIndexed DataFrame(指定索引的选择列)的以下行为感到有些困惑.

In [57]: metals
Out[57]: 
<class 'pandas.core.frame.DataFrame'>
MultiIndex: 24245 entries, (u'BI', u'Arsenic, Dissolved', -2083768576.0, 1.0) 
                        to (u'WC', u'Zinc, Total',         1661183104.0, 114.0)
Data columns:
Inflow_val      20648  non-null values
Outflow_val     20590  non-null values
Inflow_qual     20648  non-null values
Outflow_qual    20590  non-null values
dtypes: float64(2), object(2)

In [58]: metals.ix['BI'].shape  # first column in the index, ok
Out[58]: (3368, 4)

In [59]: metals.ix['BI', :, :, :].shape  # first + other columns, ok
Out[59]: (3368, 4)

In [60]: metals.ix['BI', 'Arsenic, Dissolved'].shape # first two cols
Out[60]: (225, 4)

In [61]: metals.ix['BI', 'Arsenic, Dissolved', :, :].shape # first two + all others
---------------------------------------------------------------------------
KeyError                                  Traceback (most recent call last)
<ipython-input-62-1fb577ec32fa> in <module>()
----> 1 metals.ix['BI', 'Arsenic, Dissolved', :, :].shape                              
# traceback spaghetti snipped
KeyError: 'no item named Arsenic, Dissolved'

In [62]: metals.ix['BI', 'Arsenic, Dissolved', :, 1.0].shape # also fails

我花了很长时间才意识到我一直试图实现的目标In [61]是可能的In [60].为什么该ix方法的行为如下？我真正想要的是那个场景In [62].

我的猜测是我需要重新定义索引层次结构,但我很好奇是否有更简单的方法.

谢谢.

Answer 1

如果你想根据MultiIndex级别值选择行/列,我建议使用'.xs()'方法.另请参阅使用复合(分层)索引从Pandas数据框中选择行

对于此示例,您可以使用:

#short hand:
metals.xs('BI', level=0).xs('Arsenic, Dissolved', level=0).xs(1, level=1)

# more verbose
metals.xs('BI', level='bmp_category').xs('Arsenic, Dissolved', level='parameter').xs(1, level='storm')

# two chained `ix` calls:
metals.ix['BI', 'Arsenic, Dissolved'].ix[:, 1]