KeyError: '__dummy__' for pd.crosstab in pandas

[songhuiming]

get ~~ KeyError: 'dummy' ~~ when I run the following:

np.random.seed(seed = 99)
s = np.random.randint(1,10,200)
s = pd.Series(np.where(s > 9, np.nan, s))
s1 = s[:100]
s2 = s[100:]
pd.crosstab(s1, s2)

KeyError: '__dummy__

[lexual]

Even simpler example. Perhaps something to do with the indices not overlapping at all.

s1 = pd.Series([1, 2, 3], index=[1, 2, 3])
s2 = pd.Series([4, 5, 6], index=[4, 5, 6])
pd.crosstab(s1, s2)

[lexual]

Believe this is the root cause of things:

http://pandas.pydata.org/pandas-docs/stable/groupby.html#na-group-handling

[lexual]

Yes http://pandas.pydata.org/pandas-docs/stable/groupby.html#na-group-handling is this cause.

Because the 2 indices have no overlapping indexes, this means that each groupby ends up including a nan which then excludes it from groupby result.

You then end up with an empty dataframe and that is the cause of the KeyError, as you're accessing df['dummy'] on an empty dataframe.

[jreback]

yeh, this should just be an empty frame, as there are no cross-tabulations.

[lexual]

So this is not a bug?

should we:

• raise exception
• return an empty dataframe?

[jreback]

return an empty frame

[dan7davis]

I'm getting the same KeyError: 'dummy' for my grouped data.

And I'm not really sure how to fix it / what you mean by 'return an empty frame.' Care to dumb it down/show precisely what you mean?

Thanks!

[jreback]

@dan7davis this needs a fix that would return an empty frame when catching the KeyError exception raised by the example above

https://github.com/pydata/pandas/blob/master/pandas/tools/pivot.py#L151, just need something like:

try:
    table = table[values[0]]
except KeyError:
    pass

[dan7davis]

@jreback problem solved. thank you! really appreciate the alacrity

[jreback]

want to do a pull request to fix in master?

[dan7davis]

I'm (very) new to coding/python/GitHub, so unfortunately I have no idea
what that means. But it sounds useful for me to know & helpful for others,
so I'd be happy to learn/try..

On Tue, Jan 5, 2016 at 3:32 PM, Jeff Reback notifications@github.com
wrote:

want to do a pull request to fix in master?

—
Reply to this email directly or view it on GitHub
pydata#10291 (comment).

[jreback]

contributing is a great way to learn ...., see our docs: http://pandas.pydata.org/pandas-docs/stable/contributing.html

any questions, pls ask.