BUG: (iloc) Comparisons of Series have different indexing semantics from DataFrame comparisons

[mairas]

Consider the following:

In [3]: df = pd.DataFrame(np.arange(5), columns=["foo"])

In [4]: df["foo"].iloc[:-1] != df["foo"].iloc[1:]
Out[4]:
0    True
1    True
2    True
3    True
dtype: bool

In [5]: df["foo"].iloc[1:] != df["foo"].iloc[:-1]
Out[5]:
1    True
2    True
3    True
4    True
dtype: bool

Obviously, the ne operator uses positional indexing when comparing the values and applies the index of the first argument to the result. I don't think this is the expected behaviour.

Again, the semantics are different when performing addition:

In [6]: df["foo"].iloc[:-1] + df["foo"].iloc[1:]
Out[6]:
0   NaN
1     2
2     4
3     6
4   NaN
dtype: float64

In [7]: df["foo"].values[:-1] + df["foo"].values[1:]
Out[7]: array([1, 3, 5, 7])

Here, everything works as expected.

I also tried the dataframe ne method:

In [8]: df.iloc[:-1].ne(df.iloc[1:])
Out[8]:
     foo
0   True
1  False
2  False
3  False
4   True

This works as expected, and differently from Series comparisons. The respective method doesn't exist for Series objects, so I couldn't test that.

[hayd]

Also, weirdly, if you use != rather than ne on DataFrame you get:

df.iloc[:-1] != df.iloc[1:]
Exception: Can only compare identically-labeled DataFrame objects

this behaviour may change after #3482 is merged, cc #4324 @jreback

[jreback]

I think a bug in iloc/Series impl. (they are separate for efficiency) so this case not tested

In [12]: df["foo"].iloc[1:] != df["foo"].iloc[:-1]
Out[12]: 
1    True
2    True
3    True
4    True
dtype: bool

In [13]: df["foo"].iloc[1:] == df["foo"].iloc[:-1]
Out[13]: 
1    False
2    False
3    False
4    False
dtype: bool

In [14]: df['foo'].iloc[1:]
Out[14]: 
1    1
2    2
3    3
4    4
dtype: int64

In [15]: df['foo'].iloc[-1:]
Out[15]: 
4    4
dtype: int64

Works for frames though

In [17]: df.iloc[1:] == df.iloc[4]
Out[17]: 
     foo
1  False
2  False
3  False
4   True

[jreback]

I think this is a bug in ne actually; it should raise (and Series should raise)