Int64Index with dtype=float and slicing issues

[josteinbf]

I have stumbled onto problems with slicing a pandas DataFrame when an index that originally contained integers has been manipulated such that it contains floats. Consider the following example (somewhat contrived to make it stand on its own; in practice, data are loaded from file):

df = pd.DataFrame({'i': np.linspace(0, 5, 6).astype(np.int64), 
               'j': np.linspace(2, 3, 6)})
df.set_index('i', inplace=True)
print(df.index)   # Int64Index([0, 1, 2, 3, 4, 5], dtype='int64')
df.index = df.index / 10.
print(df.index)   # Int64Index([0.0, 0.1, 0.2, 0.3, 0.4, 0.5], dtype='float64')
df[:0.3]

The final line raises this exception:

TypeError: the slice stop value [None] is not a proper indexer 
for this index type (Int64Index)

While it was relatively easy to work around this using an explicit cast along the lines of

df.index = np.asarray(df.index, dtype=np.float64) / 10.

it took me quite a while to figure out what was going on. What I expected to happen in the original code was that the index would become a Float64Index or something similar. When I first noticed the Int64Index with dtype=float I thought that was wierd. Is it a good idea to change this behaviour so that there is not this mismatch between index type and dtype?

My pandas version is 0.15.2. User EdChum at StackOverflow[1] has kindly tested this on 0.16, finding the same behaviour with the code above. Slicing with .ix works on 0.16 but not on 0.15.2; slicing with .loc works for both versions.

[1] http://stackoverflow.com/questions/29795225/pandas-int64index-with-dtype-float

[jreback]

This is a bug (but not with the slicing, rather the index construction).

This should be return Index(op(values, other), **self._get_attributes_dict()

(I think the unary ops are ok, e.g. they will always be the same type, but could change that as well).

These are not really tested all that well, so need a comprehensive test.

Interested in doing a PR?

[josteinbf]

@jreback I'm interested in doing a PR, but as I have never done any work on pandas before, I'm unsure if I will be able to. I hope it's ok for me to take a couple of days to look into it to see if I'm able to understand the code?

Also, to make sure I understand correctly: We should fix the issue that Int64Index can sometimes contain floats by making sure we create a Float64Index where appropriate. As a consequence of that fix the slicing issues I found should go away, right?

[jreback]

see contributing guidelines

The issue is that certain numeric ops return the wrong index type; all else follows from there.