Pandas dataframe.value_counts()返回浮点数索引的键

Question

我正在尝试使用pandas来获取值的计数。当我输入以下命令时：

my_variable.value_counts().keys()

我得到了如下输出：

Index([1.0, 0.0, 1.00999999046, 2.0, 2.00999999046, 3.0, 1.01000022888, 3.00999999046, 4.00999999046, 4.0, 6.00999999046, 5.00999999046, 8.00999999046, 2.01000022888, 5.0, 0.990000009537, 9.00999999046, 6.0, 7.0, 12.0099999905, 7.00999999046, 10.0099999905, 3.01000022888, 19.0199999809, 11.0099999905, 20.0199999809, 8.0, 14.0199999809, 4.01000022888, 5.01000022888, 38.0399999619, 46.0499999523, 40.0399999619, 20.0299999714, 16.0199999809, 18.0299999714, 9.01999998093, 11.0199999809, 21.0199999809, -10651.4099998, -4643.13999987, -6388.92000008, -5779.98000002], dtype=object)

我的问题是，如何访问由浮点数值组成的键，比如键1.00999999046？

我可以通过以下方式访问索引1.0：

my_variable.value_counts()[1]

但是，如果我尝试：

my_variable.value_counts()[1.00999999046]

那么我会收到一个错误提示：

KeyError: 1.00999999046

我觉得这可能和索引的类型是object有关，但我不知道该怎么处理。任何建议都将不胜感激。

Answer 1

你可以用 value_counts().index[i] 来代替。这里的 i 是你想要调用的索引的编号。

Answer 2

在版本0.13及以上，这个功能运行得很好。在0.13之前，浮点数索引并没有什么特别之处。现在它们有了新的逻辑，可以避免将索引值四舍五入或截断成整数。换句话说，浮点数的值会直接被查找，而不会被强制转换（对于Float64Index来说）。实际上，这种索引类型的目的就是为了创建一个统一的索引模型，让使用[],ix,loc时返回的结果完全一致。

可以查看文档了解更多信息。

In [8]: i = Index([1.0, 0.0, 1.00999999046, 2.0, 2.00999999046, 3.0, 1.01000022888, 3.00999999046, 4.00999999046, 4.0, 6.00999999046, 5.00999999046, 8.00999999046, 2.01000022888, 5.0, 0.990000009537, 9.00999999046, 6.0, 7.0, 12.0099999905, 7.00999999046, 10.0099999905, 3.01000022888, 19.0199999809, 11.0099999905, 20.0199999809, 8.0, 14.0199999809, 4.01000022888, 5.01000022888, 38.0399999619, 46.0499999523, 40.0399999619, 20.0299999714, 16.0199999809, 18.0299999714, 9.01999998093, 11.0199999809, 21.0199999809, -10651.4099998, -4643.13999987, -6388.92000008, -5779.98000002])

In [9]: i
Out[9]: Float64Index([1.0, 0.0, 1.00999999046, 2.0, 2.00999999046, 3.0, 1.01000022888, 3.00999999046, 4.00999999046, 4.0, 6.00999999046, 5.00999999046, 8.00999999046, 2.01000022888, 5.0, 0.990000009537, 9.00999999046, 6.0, 7.0, 12.0099999905, 7.00999999046, 10.0099999905, 3.01000022888, 19.0199999809, 11.0099999905, 20.0199999809, 8.0, 14.0199999809, 4.01000022888, 5.01000022888, 38.0399999619, 46.0499999523, 40.0399999619, 20.0299999714, 16.0199999809, 18.0299999714, 9.01999998093, 11.0199999809, 21.0199999809, -10651.4099998, -4643.13999987, -6388.92000008, -5779.98000002], dtype='object')

In [10]: s = Series(i.tolist() * 3)


In [13]: s.value_counts()[1.00999999046]
Out[13]: 3

请注意，索引的显示是截断后的值（它们实际上是完整存在的，只是在这里不显示超过两位的小数）。

In [14]: s.value_counts().sort_index()
Out[14]: 
-10651.41    3
-6388.92     3
-5779.98     3
-4643.14     3
 0.00        3
 0.99        3
 1.00        3
 1.01        3
 1.01        3
 2.00        3
 2.01        3
 2.01        3
 3.00        3
 3.01        3
 3.01        3
 4.00        3
 4.01        3
 4.01        3
 5.00        3
 5.01        3
 5.01        3
 6.00        3
 6.01        3
 7.00        3
 7.01        3
 8.00        3
 8.01        3
 9.01        3
 9.02        3
 10.01       3
 11.01       3
 11.02       3
 12.01       3
 14.02       3
 16.02       3
 18.03       3
 19.02       3
 20.02       3
 20.03       3
 21.02       3
 38.04       3
 40.04       3
 46.05       3
dtype: int64

In [15]: s.value_counts()[1.00999999046]
Out[15]: 3

In [16]: s.value_counts().keys()
Out[16]: Float64Index([3.00999999046, 14.0199999809, 2.00999999046, -10651.4099998, 2.01000022888, 18.0299999714, 20.0299999714, 16.0199999809, 6.00999999046, 3.01000022888, 8.0, 11.0199999809, 19.0199999809, 7.0, 1.01000022888, 0.990000009537, 4.0, 3.0, 2.0, 1.0, 46.0499999523, 11.0099999905, 12.0099999905, 4.00999999046, 40.0399999619, 7.00999999046, 9.01999998093, 6.0, -6388.92000008, 21.0199999809, 38.0399999619, 5.0, 20.0199999809, 4.01000022888, -5779.98000002, 1.00999999046, 9.00999999046, -4643.13999987, 5.01000022888, 10.0099999905, 8.00999999046, 5.00999999046, 0.0], dtype='object')