BUG/ENH: Bad columns dtype when creating empty DataFrame

[lowerthansound]

Code Sample

>>> df = pd.DataFrame(columns=list('ABC'), dtype='int64')
>>> df
Empty DataFrame
Columns: [A, B, C]
Index: []
>>> df.dtypes
A    float64
B    float64
C    float64
dtype: object

Problem description

When creating a DataFrame with no rows, the presence of a dtype argument may convert the columns into float64. The problem does not happen if the DataFrame has one or more rows:

>>> df = pd.DataFrame([[1, 2, 3]], columns=list('ABC'), dtype='int64')
>>> df
   A  B  C
0  1  2  3
>>> df.dtypes
A    int64
B    int64
C    int64
dtype: object

Expected Output

>>> df = pd.DataFrame(columns=list('ABC'), dtype='int64')
>>> df.dtypes
A    int64
B    int64
C    int64
dtype: object

Output of pd.show_versions()

commit: None
python: 3.6.6.final.0
python-bits: 64
OS: Linux
OS-release: 4.18.5-arch1-1-ARCH
machine: x86_64
processor:
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8

pandas: 0.23.4
pytest: 3.8.0
pip: 10.0.1
setuptools: 40.2.0
Cython: 0.28.5
numpy: 1.15.1
scipy: 1.1.0
pyarrow: 0.9.0
xarray: 0.10.8
IPython: 6.5.0
sphinx: 1.7.9
patsy: 0.5.0
dateutil: 2.7.3
pytz: 2018.5
blosc: None
bottleneck: 1.2.1
tables: 3.4.4
numexpr: 2.6.8
feather: 0.4.0
matplotlib: 2.2.3
openpyxl: 2.5.5
xlrd: 1.1.0
xlwt: 1.3.0
xlsxwriter: 1.1.0
lxml: 4.2.5
bs4: 4.6.3
html5lib: 1.0.1
sqlalchemy: 1.2.11
pymysql: 0.9.2
psycopg2: None
jinja2: 2.10
s3fs: 0.1.6
fastparquet: 0.1.6
pandas_gbq: None
pandas_datareader: None

[JustinZhengBC]

This seems to be intended behaviour, as demonstrated by the following test in pandas/tests/frame/test_constructor.py::TestDataFramConstructors::test_constructor_corner

df = DataFrame(index=lrange(10), columns=['a', 'b'], dtype=int)
    assert df.values.dtype == np.dtype('float64')

The code responsible for this behaviour is found in pandas/core/dtypes/cast.py, on line 1223. Commenting out these two lines causes the above test, and no others, to fail in the pytest suite.

if is_integer_dtype(dtype) and isna(value):
    dtype = np.float64

[lowerthansound]

I don't feel this is intended behavior, but it may be a rough corner produced by the code you mentioned.

In the issue sample, the columns are empty, therefore, no need to upcast to float:

>>> df = pd.DataFrame(columns=list('ABC'), dtype='int64')
>>> df
Empty DataFrame
Columns: [A, B, C]
Index: []

In the test case you mentioned, though, the DataFrame must be filled with NaN and therefore float is needed:

>>> df = pd.DataFrame(index=range(10), columns=['a', 'b'], dtype=int)
>>> df
    a   b
0 NaN NaN
1 NaN NaN
2 NaN NaN
3 NaN NaN
4 NaN NaN
5 NaN NaN
6 NaN NaN
7 NaN NaN
8 NaN NaN
9 NaN NaN

[JustinZhengBC]

Good point. Theoretically it could be fixed by making the int cast to float only if an lrange is specified. I can try it out later and submit a PR if the tests pass.