DataFrame.where() fails with some binary comparisons

[JohnNapier]

Suppose we want to cap values in a dataframe of shape (6,5).

import pandas as pd,numpy as np
# WORKS WITH DF OF SHAPE (6,5)
ix=pd.date_range('1/1/2015',periods=6,freq='D')
df0=pd.DataFrame(np.random.normal(size=(len(ix),5)),index=ix)
print df0
df1=df0.quantile(q=.5,axis=1)
print df1
df2=df0.where(df0<df1,df1,axis=0)
print df2

The above code has capped values above the median. For example:

                   0         1         2         3         4
2015-01-01  1.209706  1.362493 -0.141184 -0.262279  0.108538
2015-01-02 -0.752241  0.554342 -0.053927 -1.150906 -0.993721
2015-01-03  0.049342  1.100449  0.616066 -1.453542 -0.023043
2015-01-04 -1.148023 -0.438199 -1.378162  0.678946 -0.224544
2015-01-05  0.395681 -1.237518 -0.606024  0.834100  0.545389
2015-01-06 -0.426188 -0.283056 -0.012219  0.510134  1.363700

2015-01-01    0.108538
2015-01-02   -0.752241
2015-01-03    0.049342
2015-01-04   -0.438199
2015-01-05    0.395681
2015-01-06   -0.012219
Freq: D, dtype: float64

                   0         1         2         3         4
2015-01-01  0.108538  0.108538 -0.141184 -0.262279  0.108538
2015-01-02 -0.752241 -0.752241 -0.752241 -1.150906 -0.993721
2015-01-03  0.049342  0.049342  0.049342 -1.453542 -0.023043
2015-01-04 -1.148023 -0.438199 -1.378162 -0.438199 -0.438199
2015-01-05  0.395681 -1.237518 -0.606024  0.395681  0.395681
2015-01-06 -0.426188 -0.283056 -0.012219 -0.012219 -0.012219

Now, if we do the same with a squared dataframe, the same code will fail.

# FAILS WITH DF OF SHAPE (5,5)
df0=df0.iloc[:5,:5]
df1=df1.iloc[:5]
df2=df0.where(df0<df1,df1,axis=0)
print df2

For some reason, broadcasting now is done through axis=1, although the argument explicitly asked for axis=0.

                   0         1         2         3         4
2015-01-01  0.108538 -0.752241 -0.141184 -0.262279  0.395681
2015-01-02  0.108538 -0.752241  0.049342 -1.150906 -0.993721
2015-01-03  0.108538 -0.752241  0.049342 -1.453542 -0.023043
2015-01-04 -1.148023 -0.752241 -1.378162 -0.438199  0.395681
2015-01-05  0.108538 -1.237518 -0.606024 -0.438199  0.395681

The same failure will take place is we use binary comparison functions.

# IT ALSO FAILS WITH BINARY COMPARISONS
# http://pandas.pydata.org/pandas-docs/stable/basics.html#basics-compare
df2=df0.where(df0.lt(df1,axis=0),df1,axis=0)
print df2

                   0         1         2         3         4
2015-01-01  0.108538 -0.752241 -0.141184 -0.262279  0.395681
2015-01-02  0.108538 -0.752241  0.049342 -1.150906 -0.993721
2015-01-03  0.108538 -0.752241  0.049342 -1.453542 -0.023043
2015-01-04 -1.148023 -0.752241 -1.378162 -0.438199  0.395681
2015-01-05  0.108538 -1.237518 -0.606024 -0.438199  0.395681

pandas version 0.16.2 and numpy 1.9.2

Thanks!

[jreback]

should be closed by this: #9838

when posting an example like this, pls use np.random.seed(...) to make it reproducible
by copy-pasting

[jreback]

issue is this: #9736