Integers converted to floats when round-tripping through Excel

[jtratner]

0.12 and current master have this problem (problem is that we were only testing frames with floats):

>>> df = DataFrame(range(10))
>>> df.dtypes
0    int64
dtype: object
>>> df.index
Int64Index([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], dtype=int64)
>>> df.to_excel('test1.xlsx')
>>> new_df = read_excel('test1.xlsx', 'sheet1')
>>> new_df.index
Index([0.0, 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0], dtype=object)
>>> new_df.dtypes
0    float64
dtype: object

Can we do some type inference here or something? read_csv handles it better:

>>> df.to_csv('out.csv')
>>> new_df = read_csv('out.csv')
>>> new_df.index
Int64Index([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], dtype=int64)
>>> new_df.dtypes
Unnamed: 0    int64
0             int64
dtype: object

[jreback]

try this: df._data.downcast(dtype='infer')

[jtratner]

@jreback so could we add a keyword argument infer and default it to True so it roundtrips without issue? How do other parts of the library handle it?

[jreback]

you have float to int inference and int/float say to timedelta (but leave this for later)

downcast will do these 4 u (timedelta is a bit trickier)

[jreback]

need an update on this issue.