.ix[] inconsistently assigns right hand side

[etyurin]

This is with pandas 0.12.0.dev-be25266:

In [1]: import pandas as pd

In [2]: df = pd.DataFrame({'a':[1,2,3],'b':[0,1,2]})

In [3]: df.ix[[0,2,],'b'] = [100,-100] # Works as intended

In [4]: print df
a b
0 1 100
1 2 1
2 3 -100

In [6]: df = pd.DataFrame({'a':range(4)})

In [7]: df['b']=pd.np.nan

In [8]: df.ix[[1,3],'b'] = [100,-100] # Both rows get assigned 100

In [9]: print df
a b
0 0 NaN
1 1 100
2 2 NaN
3 3 100

In [10]: df['b'].ix[[1,3]] = [100,-100] # Works as intended

In [11]: print df
a b
0 0 NaN
1 1 100
2 2 NaN
3 3 -100

[jreback]

you have a list on the rhs, so it MUST be assigned by position, you are MUCH better off using a series here (which exactly specify what the values align to)

I'll look and see if there is a bug in any event

[jreback]

This is from current master (FYI I did push a change about an hour ago, but had to do with Panel alignment)

Looks ok to me

In [45]: df = pd.DataFrame({'a':[1,2,3],'b':[0,1,2]})

In [46]: df['b']=pd.np.nan

In [47]: df
Out[47]: 
   a   b
0  1 NaN
1  2 NaN
2  3 NaN

In [48]: df.ix[[0,2,],'b'] = [100,-100]

In [49]: df
Out[49]: 
   a    b
0  1  100
1  2  NaN
2  3 -100

In [50]: df = pd.DataFrame({'a':range(4)})

In [51]: df['b']=pd.np.nan

In [52]: df
Out[52]: 
   a   b
0  0 NaN
1  1 NaN
2  2 NaN
3  3 NaN

In [53]: df.ix[[1,3],'b'] = [100,-100]

In [54]: df
Out[54]: 
   a    b
0  0  NaN
1  1  100
2  2  NaN
3  3 -100

In [55]: pd.__version__
Out[55]: '0.12.0-141-g2e09fcc'

Also your very last assignment
df['b'].ix[[1,3]] = [100,-100]

You should NEVER do this; it may accidently work, but its not a guarantee (and its very hard to warn aboutt his); this is a chained assignment

[etyurin]

Thank you very much - I guess it was a version-specific issue and it's best to update to the latest.

I started doing "chained assignment" when I noticed the simple/combined assignment was not working properly. In my ignorance I actually thought it'd be more straightforward.

[jreback]

np

the chained assignment is a very touchy issue...everyone wants it to work, but python basically doesn't let you do in a nice clean way (well it doesn, but it doesn't let you have control of it!)

and its insidous because it can work (I may have completely fixed it, but not 100% sure, so that we dont' official allow it) - and neither does numpy....

[jreback]

closing as already in master (tests are in #4509)