Problem when setting values based on MultiIndex subset

[gerigk]

from pandas import *
test = read_csv('/home/arthur/transform_issue.csv')
x =test.groupby(['A','B','C'])['revenues'].first().index
test.set_index(['A','B','C'], inplace=True)
test.ix[x]['revenues']= 999.99

print test.ix[x].shape
print test[test.revenues==999.99]

(127, 5)
Empty DataFrame
Columns: array([D, E, week, revenues, orders], dtype=object)
Index: array([], dtype=object)

I then tried setting via df[col][indexes] which simply crashed my session without any exception

test = read_csv('/home/arthur/transform_issue.csv')
x =test.groupby(['A','B','C'])['revenues'].first().index
test.set_index(['A','B','C'], inplace=True)
test['revenues'][x]= 999.99

print test.ix[x].shape
print test[test.revenues==999.99]

I sent the csv via email to wes@lambdafoundry.com

[gerigk]

one more way failing:

from pandas import *
test = read_csv('/home/arthur/transform_issue.csv')
x =test.groupby(['A','B','C'])['revenues'].first().index
test.set_index(['A','B','C'], inplace=True)
test.ix[x, 'revenues']= 999.99

---------------------------------------------------------------------------
IndexError                                Traceback (most recent call last)
<ipython-input-5-62f629907fd1> in <module>()
      3 x =test.groupby(['A','B','C'])['revenues'].first().index
      4 test.set_index(['A','B','C'], inplace=True)
----> 5 test.ix[x, 'revenues']= 999.99

/usr/local/lib/python2.7/dist-packages/pandas-0.8.0rc2-py2.7-linux-x86_64.egg/pandas/core/indexing.pyc in __setitem__(self, key, value)
     62                 raise IndexingError('only tuples of length <= %d supported',
     63                                     self.ndim)
---> 64             indexer = self._convert_tuple(key)
     65         else:
     66             indexer = self._convert_to_indexer(key)

/usr/local/lib/python2.7/dist-packages/pandas-0.8.0rc2-py2.7-linux-x86_64.egg/pandas/core/indexing.pyc in _convert_tuple(self, key)
     71         keyidx = []
     72         for i, k in enumerate(key):
---> 73             idx = self._convert_to_indexer(k, axis=i)
     74             keyidx.append(idx)
     75         return tuple(keyidx)

/usr/local/lib/python2.7/dist-packages/pandas-0.8.0rc2-py2.7-linux-x86_64.egg/pandas/core/indexing.pyc in _convert_to_indexer(self, obj, axis)
    370                 # this is not the most robust, but...
    371                 if (isinstance(labels, MultiIndex) and
--> 372                     not isinstance(objarr[0], tuple)):
    373                     level = 0
    374                     _, indexer = labels.reindex(objarr, level=level)

IndexError: index 0 is out of bounds for axis 0 with size 0

[gerigk]

In case anybody wonders what that would be good for:

Say I have Gender, Date, Value as columns

I want to compute the pct_change of value and unfortunately grouped.agg(lambda x: x.pct_change() ) is very slow.
I expect calling
df['delta'] = df.values.pct_change()
ind = df.groupby('Gender').first().index
df.ix[ind, 'delta'] = np.nan
to be faster.
Since groupby kicks out dates without observations of "value" this is the smartest solution I found (otherwise I could set df.delta[df.date = min(df.date)]= np.nan).

[wesm]

I'll take a look, thanks

[wesm]

Thanks for the report, fixed these issues which were caused by some recent internal changes in MultiIndex

[gerigk]

from pandas import *
test = read_csv('/home/arthur/transform_issue.csv')
x =test.groupby(['A','B','C'])['revenues'].first().index
test.set_index(['A','B','C'], inplace=True)
test.ix[x]['revenues']= 999.99

print test.ix[x].shape
print test[test.revenues==999.99]

still doesn't work.

the other 2 ways of assigning are working now. is this way supposed to work, too? there is no exception/warning raised.

[wesm]

That won't work because test.ix[x] produces a copy