BUG: Assigning Panel item using .ix/.loc no longer works

[creeson]

I am updating a codebase from an OLD version of pandas (0.7.3) to current version (0.14.1).

pandas: 0.14.1
numpy: 1.8.0

The following code has stopped working correctly (using .ix[] or .loc[]):

import numpy as np
import pandas as ps

np.random.seed(0)

index=range(3)
columns = list('abc')

panel = ps.Panel({'A' : ps.DataFrame(np.random.randn(3, 3), index=index, columns=columns),
                  'B' : ps.DataFrame(np.random.randn(3, 3), index=index, columns=columns),
                  'C' : ps.DataFrame(np.random.randn(3, 3), index=index, columns=columns)
                  })

replace = ps.DataFrame(np.eye(3,3), index=range(3), columns=columns)

print panel['A']

for idx in list('ABC'):
    panel.loc[idx,:,:] = replace

print panel['A']

Output:

          a         b         c
0  1.764052  0.400157  0.978738
1  2.240893  1.867558 -0.977278
2  0.950088 -0.151357 -0.103219
    a   b   c
0 NaN NaN NaN
1 NaN NaN NaN
2 NaN NaN NaN

However, if I simply do this:

for idx in list('ABC'):
    panel[idx] = replace

print panel['A']

It behaves as expected:

   a  b  c
0  1  0  0
1  0  1  0
2  0  0  1

Also, if I assign more than just the item, it works fine:

panel.loc['A', 0, 'a'] = 1.0
print panel['A']

          a         b         c
0  1.000000  0.400157  0.978738
1  2.240893  1.867558 -0.977278
2  0.950088 -0.151357 -0.103219

Thanks.

[jreback]

hmm, must be an alignment issue. thanks, marking as a bug.

note that

panel.loc['A'] = replace works as well. Its when specifiying multiple indexers it
gets confused.

[spitz-dan-l]

Any update on this? I have a situation where I need to replace the columns of a 'minor'-oriented panel, necessitating p.loc[:, :, idx] = replace

Any suggested workarounds for this?

[jreback]

that syntax should work
give an example is it does not

[spitz-dan-l]

Investigation showed what's different about my use case. My issue is I am trying to add columns. For the example above, my panel was set up like so:

panel = ps.Panel.from_dict({
                  'A' : ps.DataFrame(np.random.randn(3, 3), index=index, columns=columns),
                  'B' : ps.DataFrame(np.random.randn(3, 3), index=index, columns=columns),
                  'C' : ps.DataFrame(np.random.randn(3, 3), index=index, columns=columns)
                  }, orient='minor')

I'm trying to add a 'd' column to the panel:

replace = ps.DataFrame(np.eye(4,4), index=range(4), columns=list('abcd'))

such that only the 'A' label gets its values replaced. For B and C, the 'd' column should be NaN. The way I'd thought to do it was:

panel.loc[:, :, 'A'] = replace

But this doesn't add 'd'. I end up with:

print panel.loc[:, :, 'A']

   a  b  c
0  1  0  0
1  0  1  0
2  0  0  1

Is this how it ought to behave? Am I asking for what I want the wrong way?

[spitz-dan-l]

@jreback yes. Example code: https://gist.github.com/spitz-dan-l/01a8e7788ffaeec87fd7

[jreback]

@spitz-dan-l
the indicated behavior is correct; the panel will reindex the provide data to the same as the panel.

prob best to just reindex the axes to the union of the axes you want

e.g. something like

panel = panel.reindex(items=panel.items|replace.columns,major_axis=panel.major_axis|replace.index)
panel.loc[:,:,'A'] = replace

or you can use concat and update. .loc is not the right tool here because of the ambiguity. pandas normally aligns the input (special case on the 0th axis where it can enlarge).

[spitz-dan-l]

thanks!