BUG/ENH: dtype change on enlargement

[jreback]

http://stackoverflow.com/questions/22044766/adding-row-to-pandas-dataframe-changes-dtype/22045011#22045011

A bit non-trival (maybe coerce back after) if possible?

maybe just change this to an actual append?

In [1]: df = DataFrame({'a' : range(10)}, dtype='i4')

In [2]: df
Out[2]: 
   a
0  0
1  1
2  2
3  3
4  4
5  5
6  6
7  7
8  8
9  9

[10 rows x 1 columns]

In [3]: df.dtypes
Out[3]: 
a    int32
dtype: object

In [4]: df.loc[10] = 99

In [5]: df
Out[5]: 
     a
0    0
1    1
2    2
3    3
4    4
5    5
6    6
7    7
8    8
9    9
10  99

[11 rows x 1 columns]

In [6]: df.dtypes
Out[6]: 
a    float64
dtype: object

[goretkin]

Is there a way to just assert the type on an enlargement?

I'm trying to avoid this...

In [1]: import pandas as pd

In [2]: names_df = pd.DataFrame(
   ...:             {"id":pd.Series([], dtype='int64')
   ...:              })
   ...: names_df.dtypes
   ...: 
Out[2]: 
id    int64
dtype: object

In [3]: names_df.loc[0, "id"] = 3
   ...: names_df.dtypes
   ...: 
Out[3]: 
id    float64
dtype: object

[jorisvandenbossche]

You can avoid this by not assigning with enlargement.
Options are populating the dataframe on construction, or using concat

[mroeschke]

Adding to a non-empty dataframe no longer enlarges (Jeff's example) but the empty case still does.

[mroeschke]

The first example could use a regression test since it works on master.

In [49]: In [1]: df = DataFrame({'a' : range(10)}, dtype='i4')
    ...:

In [50]: df
Out[50]:
   a
0  0
1  1
2  2
3  3
4  4
5  5
6  6
7  7
8  8
9  9

In [51]: df.dtypes
Out[51]:
a    int32
dtype: object

In [52]: In [4]: df.loc[10] = 99
    ...:

In [53]: df.dtypes
Out[53]:
a    int64
dtype: object

In [56]: pd.__version__
Out[56]: '0.26.0.dev0+627.gef77b5700'

[mroeschke]

Noted #30444 (comment): The first example should maintain the int32 typing so this is not completely fixed.