Original numpy array getting altered when changes made to dataframe

[VathsalaAchar]

Code Sample

a = np.array([1,2,3, np.nan])
b = pd.DataFrame(a)
b.fillna(4, inplace=True)
print b
print a

Output

     0
0  1.0
1  2.0
2  3.0
3  4.0
[ 1.  2.  3.  4.]

Problem description

When a dataframe is created from a numpy array the changes to the dataframe are altering the original numpy array. I did not expect this to happen and I'm not sure if this is an expected behaviour or a known issue.

I do know how to work around this, but my question is whether I have to.

Expected Output

     0
0  1.0
1  2.0
2  3.0
3  4.0
[  1.   2.   3.  nan]

Output of pd.show_versions()

INSTALLED VERSIONS

commit: None
python: 2.7.12.final.0
python-bits: 64
OS: Linux
OS-release: 4.4.8-040408-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_GB.UTF-8

pandas: 0.18.0
nose: 1.3.7
pip: 8.1.1
setuptools: 20.3
Cython: 0.23.4
numpy: 1.10.4
scipy: 0.17.0
statsmodels: 0.6.1
xarray: None
IPython: 4.1.2
sphinx: 1.3.5
patsy: 0.4.0
dateutil: 2.5.1
pytz: 2016.6.1
blosc: None
bottleneck: 1.0.0
tables: 3.2.2
numexpr: 2.5
matplotlib: 1.5.1
openpyxl: 2.3.2
xlrd: 0.9.4
xlwt: 1.0.0
xlsxwriter: 0.8.4
lxml: 3.6.0
bs4: 4.4.1
html5lib: None
httplib2: None
apiclient: None
sqlalchemy: 1.0.12
pymysql: None
psycopg2: None
jinja2: 2.8
boto: 2.40.0

[jreback]

so this is a 'feature', in that view propogation in numpy is a feature. As a user you have to be congnizant of it, and it can make things quite performant. Pandas does not own a passed in numpy array and thus it IS externally visible.

In general, using inplace=True ops are not idiomatic to pandas, virtually all operations return new data (which is copied).

Note that view propogation is only true in some cases: single dtyped, no prior modification, no dtype changes on the op, and non-object types.

In [1]: a = np.array([1,2,3, np.nan])
   ...: b = pd.DataFrame(a)
   ...: b.fillna(4, inplace=True)

# this is the viewed array
In [2]: b.values.base
Out[2]: array([ 1.,  2.,  3.,  4.])

In [3]: a2 = np.array(['a', 'b', 'c'])

# not true for object dtypes
In [4]: b2 = pd.DataFrame(a2)

In [5]: b2.loc[0, 0] = 'foo'

In [6]: b2
Out[6]: 
     0
0  foo
1    b
2    c

In [7]: a2
Out[7]: 
array(['a', 'b', 'c'], 
      dtype='<U1')

This happens to be only in-place in pandas itself and not numpy.

In [8]: a = np.array([1,2,3, np.nan])   ...: 

In [9]: b = pd.DataFrame(a)

In [10]: b +=1 

In [11]: a
Out[11]: array([  1.,   2.,   3.,  nan])

In [13]: b.values.base
Out[13]: array([[  2.,   3.,   4.,  nan]])