BUG: Mismatch between get and set behavior for slices of floating indices

[stephenwlin]

When df.index.inferred_type is 'floating', df.ix[start:end] uses only label-based indexing only when both start and end are floating on the getitem side but when either start or end are floating on the setitem side (where start and end are both either integers or floats within epsilon of integers)

In [2]: df=p.DataFrame(["a", "b", "c", "d"], index=[1.00, 2.00, 3.00, 4.00])

In [3]: df.ix[1.0:4] # uses positional indexing to get last three rows
Out[3]: 
   0
2  b
3  c
4  d

In [4]: df.ix[1.0:4.0] # uses label-based indexing to get all four rows
Out[4]: 
   0
1  a
2  b
3  c
4  d

In [5]: df.ix[1.0:4] = "y" # uses label-based indexing to set all four rows

In [6]: df
Out[6]:
   0
1  y
2  y
3  y
4  y

In [7]: df.ix[1.0:4.0] = "z" # uses label-based indexing to set all four rows

In [8]: df
Out[8]:
   0
1  z
2  z
3  z
4  z

[stephenwlin]

Seems like changing the getitem behavior to match setitem breaks lots of unit tests, while the reverse does not, so I will go with the latter...

Results after fix:

In [2]: df=p.DataFrame(["a", "b", "c", "d"], index=[1.00, 2.00, 3.00, 4.00])

In [3]: df.ix[1.0:4] # uses positional indexing to get last three rows
Out[3]:
   0
2  b
3  c
4  d

In [4]: df.ix[1.0:4.0] # uses label-based indexing to get all four rows
Out[4]:
   0
1  a
2  b
3  c
4  d

In [5]: df.ix[1.0:4] = "y" # uses positional indexing to set last three rows

In [6]: df
Out[6]:
   0
1  a
2  y
3  y
4  y

In [7]: df.ix[1.0:4.0] = "z" # uses label-based indexing to set all four rows

In [8]: df
Out[8]:
   0
1  z
2  z
3  z
4  z

[jreback]

did you fix this? or is this informational?

[stephenwlin]

yeah, it's "fixed" in #2777 in the sense that the get and set behaviors are consistent now, but they still don't make any sense really...

[jreback]

ahh I see
using floating point indicies is really weird anyhow
in fact probably should outlaw it :)

I mean already have label, integer, datetime