BUG: astype assignment via iloc/loc not working

[jreback]

http://stackoverflow.com/questions/17778139/pandas-unable-to-change-column-data-type/17778560#17778560

This might be trying to coerce object dtype to a real dtype (int/float) and is failing
Should prob raise for now (or work). Not working with iloc/loc.

In [66]: df = DataFrame([['1','2','3','.4',5,6.,'foo']],columns=list('ABCDEFG'))

In [67]: df.dtypes
Out[67]: 
A     object
B     object
C     object
D     object
E      int64
F    float64
G     object
dtype: object

In [68]: df.iloc[:,0:3] = df.iloc[:,0:3].astype(int)

In [69]: df.dtypes
Out[69]: 
A     object
B     object
C     object
D     object
E      int64
F    float64
G     object
dtype: object

In [70]: df.iloc[:,0:3] = df.iloc[:,0:3].convert_objects(convert_numeric=True)

In [71]: df.dtypes
Out[71]: 
A     object
B     object
C     object
D     object
E      int64
F    float64
G     object
dtype: object

[TomAugspurger]

@jreback is this fixed or did you want to raise?

In [10]: df = pd.DataFrame({'A': [1., 2., 3., 4.]})

In [11]: df.dtypes
Out[11]: 
A    float64
dtype: object

In [12]: df.loc[:, 'A'] = df['A'].astype(np.int64)

In [13]: df.dtypes
Out[13]: 
A    float64
dtype: object

In [14]: pd.__version__
Out[14]: '0.13.0rc1-27-g4d5ca5c'

I think this is still the same.

[TomAugspurger]

Ah, I guess it's just loc that isn't working. iloc handles the dtype changes fine.

[jreback]

hmm thought all those cases were tested

can u create a new issue pls

[rishik-bot]

Hello Guys,
In reference to above issue, I recently encountered something similar. In a dataframe with around 40+ columns I am trying to change dtype for first 27 columns from float to int by using iloc:
df1.iloc[:,0:27]=df1.iloc[:,0:27].astype('int')
However, it's not working. I'm not getting any error, but dtype is not changing as well.
Now the strangest part:
If I first change dtype for only 1st column (like below):
df1.iloc[:,0]=df1.iloc[:,0].astype('int')
and then run the earlier line of code:
df1.iloc[:,0:27]=df1.iloc[:,0:27].astype('int')
It works as required.
Any help to understand this will be grateful. Thanks!

[pierrelouisbescond]

Hi,
I am experiencing a similar issue:
I am starting from a 35 x 4800 DataFrame where all columns' types are "object" after a SimpleImputer.
When trying to go back to "int" on the first 30 columns that are integers:

df.iloc[:, 0:30] = df.iloc[:, 0:30].astype(int) => does not work

...

df.iloc[:, 0] = df.iloc[:, 0].astype(int)
df.iloc[:, 1:30] = df.loc[:, "first_col"].astype(int) => works

...

df.loc[:, "first_col"] = df.loc[:, "first_col"].astype(int)
df.iloc[:, 1:30] = df.loc[:, "first_col"].astype(int) => works

...

df.iloc[:, 0:30].astype(int).dtypes => shows all columns to "int32"