DataFrame.ix[n] for numerical (float) indices has an error

[james18]

So my understanding for a DataFrame with numerical indices, DataFrame.ix[i] indexes to the position where index=i, and not the ith row of the DataFrame. Although I find this to be confusing, i accept that this was the decision that was made.

However, the getting and setting with .ix is not consistent

from pandas import DataFrame
df = DataFrame(['a','b','c'],index=[0,0.5,1])
df.ix[1] = 'zoo'
print df.ix[1]

Current:
'b'

Expected:
'zoo'

Basically, df.ix[1] returns 'b', of index 0.5, instead of 'zoo' of index 1

[adamklein]

I'm not confident this is bug, just a very confusing circumstance.

The index is of type float. You are indexing by an integer, however.

So df.ix[1] is taking the integer index (offset 1). You really want df.ix[1.], which is the label index (offset 2).

In [12]: df.ix[1.]
Out[12]:
0 zoo
Name: 1.0

[adamklein]

I'll leave open to see what Wes has to say. This could be either be a nasty gotcha we should document, or could maybe be something we'd want to disallow (ie, integer indexing in float-based index)

[james18]

I think the bug is that it gets and sets different locations in the DataFrame. Why wouldn't it be consistent?

On Jan 27, 2012, at 8:24 PM, Adam Kleinreply@reply.github.com wrote:

I'll leave open to see what Wes has to say. This could be either be a nasty gotcha we should document, or could maybe be something we'd want to disallow (ie, integer indexing in float-based index)

---

Reply to this email directly or view it on GitHub:
https://github.com/wesm/pandas/issues/707#issuecomment-3697230

[adamklein]

Ah yes, this seems like a bug, the inconsistency of how it treats the .ix integer parameter on getting vs setting (in the context of a float index).

[adamklein]

latest commit should fix. great catch.