Python pandas数据帧用其他系列填充NaN

Question

I want to fill NaN values in a DataFrame (df) column (var4) based on a control table (fillna_mean) using column mean, and var1 as index.In the dataframe I want them to match on var1.

I have tried doing this with fillna but I dont get it to work all the way. How do I do this in a smart way, using df.var1 as index matching fillna_mean.var1?

df:

df = pd.DataFrame({'var1' : list('a' * 3) + list('b' * 2) + list('c' * 4) + list('d' * 3)
         ,'var2' : [i for i in range(12)]
         ,'var3' : list(np.random.randint(100, size = 12))
         ,'var4' : [1, 2, np.nan, 3, 2, np.nan, 1, 34, np.nan, np.nan, 12, 12]
     })

fillna_mean:

fillna = pd.DataFrame({'var1' : ['a', 'b', 'c', 'd'],
                       'mean' : [1, 3.5, 6.5, 10]})

End result is this:

var1 var2 var3  var4
a    0    69    1.0
a    1    17    2.0
a    2    83    1.0
b    3    12    3.0
b    4    36    2.0
c    5    68    6.5
c    6    13    1.0
c    7    30    34.0
c    8    23    6.5
d    9    82    10.0
d    10   32    12.0
d    11   19    12.0

Thanks in advance for input!

/swepab

Answer 1

you can use boolean indexing in conjunction with .map() method:

In [178]: fillna.set_index('var1', inplace=True)

In [179]: df.loc[df.var4.isnull(), 'var4'] = df.loc[df.var4.isnull(), 'var1'].map(fillna['mean'])

In [180]: df
Out[180]:
   var1  var2  var3  var4
0     a     0    40   1.0
1     a     1    97   2.0
2     a     2    34   1.0
3     b     3     6   3.0
4     b     4    19   2.0
5     c     5    47   6.5
6     c     6    65   1.0
7     c     7    29  34.0
8     c     8    48   6.5
9     d     9    88  10.0
10    d    10    40  12.0
11    d    11    23  12.0

Explanation:

In [184]: df.loc[df.var4.isnull()]
Out[184]:
  var1  var2  var3  var4
2    a     2    75   NaN
5    c     5    75   NaN
8    c     8    44   NaN
9    d     9    34   NaN

In [185]: df.loc[df.var4.isnull(), 'var1']
Out[185]:
2    a
5    c
8    c
9    d
Name: var1, dtype: object

In [186]: df.loc[df.var4.isnull(), 'var1'].map(fillna['mean'])
Out[186]:
2     1.0
5     6.5
8     6.5
9    10.0
Name: var1, dtype: float64

---

UPDATE: starting from Pandas 0.20.1 the .ix indexer is deprecated, in favor of the more strict .iloc and .loc indexers .

Answer 2

Get faster results with combine_first , and you don't bother you filter out nonnull data:

fillna.set_index('var1', inplace=True)

df.var4 = df.var4.combine_first(df.var1.map(fillna['mean']))