Failed to assign new value to masked serie

[omar-masmoudi]

Hello,

#!/bin/env python
import pandas as pd
import numpy as np
import sys
print sys.version
print "NUMPY:", np.__version__
print "PANDAS:", pd.__version__

df = pd.DataFrame([[0, None], [4, 3]], columns=['a','b'])
print df

df["c"] = 1.0
#print df #uncomment and result change

df["c"][~df.all(axis=1)] = 2
print df
print df.c

gives

2.7.2 (default, Jul 30 2012, 17:32:47)
[GCC 4.1.2 20080704 (Red Hat 4.1.2-46)]
NUMPY: 1.7.1
PANDAS: 0.12.0
   a   b
0  0 NaN
1  4   3
   a   b  c
0  0 NaN  1
1  4   3  1
0    2
1    1
Name: c, dtype: float64

As you can see, I try to change value of the c columns to "2" for the element where one of the column is null and it does not work.

The serie is however modified.
What is even stranger is that adding an intermediate print (uncomment above) changes the behavior.

I was not able to find any duplicate of this issue.
Thanks in advance for having a look.

[jreback]

this is a duplicate/related #5216. their are several issues that address various aspects of this.
In any event, this is fixed in 0.13.

the proper method of setting is

df.loc[~df.all(1),'c'] = 2 which works fine in 0.12

doing df['c'][df.all(1)]` is a chained assignment and can work at times, but is dependent on whether you have a mixed dtype frame or not and what exactly you are updating hence lways better to use ''.loc''

see here for a more detailed discussion