Concat coerces ints to floats if empty DataFrame is present

[EVaisman]

It seems that when using concat coerces Series of type int to type float if one of the DataFrames being concatted is empty:

>>> df1 = pd.DataFrame([1],columns=['a'])
>>> df2 = pd.DataFrame(columns=['a'])
>>> df = pd.concat([df1, df2])
>>> df['a'].dtype
dtype('float64')
>>> df1['a'].dtype
dtype('int64')

While if both columns have ints, no coercion happens:

>>> df1 = pd.DataFrame([1],columns=['a'])
>>> df2 = pd.DataFrame([1],columns=['a'])
>>> df = pd.concat([df1, df2])
>>> df['a'].dtype
dtype('int64')

[jreback]

cc @immerrr

This is actually somewhat of a grey area. If you had passed a None, then it would be excluded and so no coercion happens. The fact that you passed a completely empty frame, which is object dtype is the issue. Technically this should coerce, though I can see the case for 'ignoring' it. Since it doesn't have any valid shape.

In [21]: np.prod(df2.shape)
Out[21]: 0

In [22]: np.prod(df1.shape)
Out[22]: 1

IOW, I could see ignoring these in the pre-computation.

[immerrr]

I agree, it should coerce technically, but not practically.

[immerrr]

I have a lot on my plate right now, but if it waits till I'm done with that, I'll fix this.

[a-y-khan]

In recent Pandas version (v1.0.3), concat coerces ints to object instead of floats:

In [1]: import pandas as pd                    

In [2]: pd.__version__                         
Out[2]: '1.0.3'

In [3]: df1 = pd.DataFrame([1],columns=['a'])  

In [4]: df1['a'].dtype                         
Out[4]: dtype('int64')

In [5]: df2 = pd.DataFrame(columns=['a'])      

In [6]: df2['a'].dtype                         
Out[6]: dtype('O')

In [7]: df = pd.concat([df1, df2])             

In [8]: df['a'].dtype                          
Out[8]: dtype('O')

[mroeschke]

I think the above behavior was an intentional behavior change a while back (that an empty DataFrame has an object dtype`); therefore, a common dtype between object and int is object.

Closing as the new expected behavior, but happy to reopen if I am misunderstanding