在将NaN添加到熊猫系列中时，是否保留布尔值和浮动值之间的区别？

Question

I am adding data to a pandas Series via the Series#append method. Unfortunately, when nan is added to a bool Series, it is automatically converted to a float Series. Is there any way to avoid this conversion, or at least coerce it to object dtype, so as to preserve the distinction between bool s and float s?

>>> Series([True])                            
0    True
dtype: bool
>>> Series([True]).append(Series([np.nan]))
0     1
0   NaN
dtype: float64

Answer 1

As @Jeff said, the best way is going to be to append a Series with object dtype

Here's an example using Series

s = Series([True])
s.append(Series([nan], index=[1], dtype=object))

yielding

0    True
1     NaN
dtype: object

And one with a DataFrame :

df = DataFrame({'a': rand(10) > 0.5, 'b': randn(10)}, columns=list('ab'))
df2 = DataFrame({'a': Series([nan], dtype=object), 'b': [1.0]}, columns=df.columns, index=[len(df)])
df3 = df.append(df2)
print df3
print
print df3.dtypes

which gives

        a      b
0   False -0.865
1    True -0.186
2    True  0.078
3    True  0.995
4   False -1.420
5    True -0.340
6    True  0.042
7    True -0.627
8    True -0.217
9    True  1.226
10    NaN  1.000

a     object
b    float64
dtype: object

It's a bit clunky looking, but if you've already got the Series then you can do s.astype(object) to convert them to object dtype before appending.