python pandas create dataframe and force multiple column types

Question

I was able to create dataframe and force one data type by

import pandas as pd
test = pd.DataFrame({'a':[1,2,3], 'b':[1.1,2.1,3.1]}, dtype=int)

But I want to specify type for each column. How can I do this? I tried the following which doesn't work as the resulting dtypes are objects and b columns are not casted into integers.

test = pd.DataFrame({'a':[1,2,3], 'b':[1.1,2.1,3.1]}, dtype=[('a', int),('b', int)])

Jeff helped with above case. But I found another problem when I try to create an empty dataframe and I want to be able to specify column types. For single type across columns, I could do

test = pd.DataFrame(columns=['a','b'], dtype=int)

What if I want to specify type for each of 'a' and 'b'?

Answer 1

You can pass in a dictionary of numpy arrays, with specified dtypes; this works for creating both filled and empty arrays. (This answer is a slight adaptation on my answer here.)

Here's an empty array:

df = pd.DataFrame(data={'a' : np.array([], dtype=int),
                        'b' : np.array([], dtype=float)
                       }
                 )

Here's a filled_array:

df = pd.DataFrame(data={'a' : np.array([1,2,3], dtype=int),
                        'b' : np.array([4,5,6], dtype=float)
                       }
                 )

And you can use basically any type for dtype, such as object, str, datetime.datetime or CrazyClassYouDefined. That said, if pandas doesn't specifically support a type (such as str), pandas will fall back to treating that column as object. Don't worry though, everything should still work.

Answer 2

You can pass in a Series which has a dtype parameter

In [15]: pd.DataFrame({'a':[1,2,3], 'b':[1.1,2.1,3.1]}).dtypes
Out[15]: 
a      int64
b    float64
dtype: object

In [16]: pd.DataFrame({'a':Series([1,2,3],dtype='int32'), 'b':Series([1.1,2.1,3.1],dtype='float32')}).dtypes
Out[16]: 
a      int32
b    float32
dtype: object