BUG: DataFrame._get_value not working for IntervalIndex

[jorisvandenbossche]

Follow-up on #27926 and #27865

DataFrame._get_value with an IntervalIndex is not working properly:

In [24]:         df = DataFrame( 
    ...:             np.random.randn(5, 2), 
    ...:             columns=["a", "b"], 
    ...:             index=pd.IntervalIndex.from_breaks([-np.inf, 0, 1, 2, 3, np.inf]), 
    ...:         )  

In [26]: df._get_value(df.index[0], 'a')
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-26-ce242a095ada> in <module>
----> 1 df._get_value(df.index[0], 'a')

~/scipy/pandas/pandas/core/frame.py in _get_value(self, index, col, takeable)
   2953 
   2954         try:
-> 2955             return engine.get_value(series._values, index)
   2956         except KeyError:
   2957             # GH 20629

AttributeError: 'pandas._libs.interval.IntervalTree' object has no attribute 'get_value'

This _get_value is used in ix which is deprecated, but also in at which is not deprecated:

In [27]: df.at[df.index[0], 'a']
...
AttributeError: 'pandas._libs.interval.IntervalTree' object has no attribute 'get_value'

[another-green]

I will give a try

[jbrockmendel]

cc @jschendel i ran into this recently, too. any idea what it would take to implement IntervalTree.get_value?

[jschendel]

I think IntervalTree.get_value would be non-trivial to implement since it'd have to work for both Interval keys and numeric keys, and the get_loc logic for this is all currently at the IntervalIndex level.

That being said, it looks like adding AttributeError to the except clause in DataFrame._get_value provides a workaround:

diff --git a/pandas/core/frame.py b/pandas/core/frame.py
index 676b78573..30bda9775 100644
--- a/pandas/core/frame.py
+++ b/pandas/core/frame.py
@@ -2898,7 +2898,7 @@ class DataFrame(NDFrame):
             if self.index.nlevels > 1:
                 # partial indexing forbidden
                 raise
-        except (TypeError, ValueError):
+        except (AttributeError, TypeError, ValueError):
             pass
 
         # we cannot handle direct indexing

With the diff applied:

In [2]: df = pd.DataFrame( 
   ...:     np.arange(9).reshape(3,3), 
   ...:     columns=list('ABC'), 
   ...:     index=pd.interval_range(0, 3) 
   ...: )

In [3]: df
Out[3]: 
        A  B  C
(0, 1]  0  1  2
(1, 2]  3  4  5
(2, 3]  6  7  8

In [4]: df.at[df.index[0], 'A']
Out[4]: 0

In [5]: df.at[df.index[1], 'B']
Out[5]: 4

In [6]: df.at[0.5, 'C']
Out[6]: 2

[jbrockmendel]

Would the issue be resolved if instead of calling index._engine.get_value we used index.get_value?

[jbrockmendel]

closed by #41846