Result of add/sub/mul/div Series and DataFrame depends on type of index

[qwhelan]

Hi,

We came across this bug while attempting to demean a DataFrame. Here's typical behavior:

aapl = DataReader('AAPL', data_source='yahoo')
aapl - aapl.mean(1)

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 251 entries, 2011-09-19 00:00:00 to 2012-09-14 00:00:00
Data columns:
Open         251  non-null values
High         251  non-null values
Low          251  non-null values
Close        251  non-null values
Volume       251  non-null values
Adj Close    251  non-null values
dtypes: float64(6)

However, if we perform this same operation with an Int64Index rather than a DatetimeIndex, we get the following:

aapl = DataReader('AAPL', data_source='yahoo')
aapl.index = range(251)
aapl - aapl.mean(1)

<class 'pandas.core.frame.DataFrame'>
Int64Index: 251 entries, 0 to 250
Columns: 257 entries, 0 to Volume
dtypes: float64(257)

In the first case we got a 251x6 array, but in the latter we get a 251x257 array. It seems there's a transpose in the logic for DatetimeIndices that is not present for other indices.

[wesm]

Yes-- this is a legacy convenience wart before the existence of the add/mul/div/sub methods:

http://pandas.pydata.org/pandas-docs/stable/dsintro.html#data-alignment-and-arithmetic

In the special case of working with time series data, if the Series is a TimeSeries 
(which it will be automatically if the index contains datetime objects), and the 
DataFrame index also contains dates, the broadcasting will be column-wise:

To get consistent cross-sectional demeaning regardless of the index type, you would want to do:

aapl.sub(aapl.mean(1), axis=0)

this means: subtract the cross-sectional mean at each time point from each column. axis=0 in this context means "align on the row index" (vs. axis=1 being align on the columns)

[wesm]

This has been deprecated in 0.10.x and will be removed in 0.11 later this year