Pandas single row indexing without casting

user_1520091 0
Viewed 336

Indexing a single row of a Pandas DataFrame containing objects (strings) returns a Series which is of type object.

df1 = pd.DataFrame({'a': np.arange(3), 'b': np.arange(0.5,3),'c': list('def')})

df1
   a    b  c
0  0  0.5  d
1  1  1.5  e
2  2  2.5  f

df1.loc[0]
a      0
b    0.5
c      d
Name: 0, dtype: object

Indexing a single row of a Pandas DataFrame not containing objects (strings) returns a Series which casts its values. Column a is of type int64 in the DataFrame. The resulting series contains the value of column a as a float64. 

df2 = pd.DataFrame({'a': np.arange(3), 'b': np.arange(0.5,3)})

df2
  a    b
0  0  0.5
1  1  1.5
2  2  2.5

df2.loc[0]

a    0.0
b    0.5
Name: 0, dtype: float64

This behavior does not seem consistant, does it? How can I select a single row of the second DataFrame df2 without casting my ints to floats?

pandas dataframe indexing python
2 Answers

When there is more than one type in the Series , pandas will using object to handle it , If it is float and int , it will change the int to float

df1.loc[0].map(type)
Out[6]: 
a      <class 'numpy.int32'>
b    <class 'numpy.float64'>
c              <class 'str'>
Name: 0, dtype: object
user_79645270

A work around solution, according to the stackoverflow link posted in the comments by @Michael:

desired_index = 3

df.astype('O').iloc[desired_index]

My dataframe is big and converting the whole DataFrame to object takes a while. Here is a faster but ulgier solution.

df.iloc[[desired_index]].astype('O').iloc[0]
user_15200910
Related