BUG: where incorrectly coerces to str

[sinhrks]

Replacing some numeric values using where affects to dtype unrelated to the condition.

s = pd.Series(np.random.randn(5))
u = s.where(s<0, 'X', axis=1)
u
#0     -1.01421184508
#1                  X
#2                  X
#3    -0.228117839063
#4    -0.393903873887
# dtype: object

The result looks OK, but internally coerced to all str.

u.apply(type)
#0    <type 'str'>
#1    <type 'str'>
#2    <type 'str'>
#3    <type 'str'>
#4    <type 'str'>
# dtype: object

[sinhrks]

This seems to occur when data passed as str dtype because of numpy spec. I'll fix this if this is unintended.

Also, found different error.

w = s.where(s>1, ['A', 'B', 'C', 'D', 'E'])
# ValueError: invalid literal for long() with base 10: 'A'

w = s.where(s>1, np.array(['A', 'B', 'C', 'D', 'E']))
# ValueError: invalid literal for long() with base 10: 'A'