Axis behaviour not consistent for df.interpolate for pad with `limit_area` passed.

[mwaddoups]

The axis along which interpolation is performed is different for different methods in interpolate.

For example

df = pd.DataFrame(np.array([
    [1, 2, 3],
    [5, np.nan, np.nan],
    [3, np.nan, 4],
]))

print(df.interpolate(method='pad', axis='index'))

print(df.interpolate(method='pad', axis='index', limit_area='inside'))

returns

     0    1    2
0  1.0  2.0  3.0
1  5.0  2.0  6.0
2  3.0  2.0  4.0

     0    1    2
0  1.0  2.0  3.0
1  5.0  5.0  6.0
2  3.0  3.0  4.0

Clearly with limit_area set the interpolation happens on the wrong axis.

Not contributed before but happy to have a go at this one?

[mzeitlin11]

@mwaddoups on current master I get

     0    1    2
0  1.0  2.0  3.0
1  5.0  2.0  3.0
2  3.0  2.0  4.0
     0    1    2
0  1.0  2.0  3.0
1  5.0  NaN  NaN
2  3.0  3.0  4.0

But agreed this output doesn't look right, it appears to be interpolating on the wrong axis like you mention:

df = pd.DataFrame(np.array([
    [1, np.nan, 3],
    [5, np.nan, np.nan],
    [3, np.nan, 4],
]))

print(df.interpolate(method='pad', axis='index', limit_area='inside'))

gives

     0    1    2
0  1.0  1.0  3.0
1  5.0  NaN  NaN
2  3.0  3.0  4.0

If you're interested in fixing this, that would be awesome! Please let us know if you run into any issues!

[debnathshoham]

Hi, is anyone working on this?

[mwaddoups]

@debnathshoham all yours.

[debnathshoham]

When the DataFrame is converted to BlockManager it automatically gets transposed.
Here below, obj is same as df, but the i can see obj._mgr.blocks[0].values to be transposed. The interpolation happens on this, so interpolation along wrong axis makes sense. I am still not able to quite figure out how BlockManager works, some advice please?

obj
     0    1    2
0  1.0  2.0  3.0
1  5.0  NaN  NaN
2  3.0  NaN  4.0

obj._mgr.blocks[0].values
array([[ 1.,  5.,  3.],
       [ 2., nan, nan],
       [ 3., nan,  4.]])

pandas/pandas/core/generic.py

Lines 6826 to 6836 in 2001798

	new_data = obj._mgr.interpolate(
	method=method,
	axis=axis,
	index=index,
	limit=limit,
	limit_direction=limit_direction,
	limit_area=limit_area,
	inplace=inplace,
	downcast=downcast,
	**kwargs,
	)