如何根据其他变量估算缺失值

Question

I have a dataframe like below:

df = pd.DataFrame({'one' : pd.Series(['a', 'b', 'c', 'd','aa','bb',np.nan,'b','c',np.nan, np.nan] ), 
  'two' : pd.Series([10, 20, 30, 40,50,60,10,20,30,40,50])} )

In which first column is the variables, second column is the values. Variable value is constant, which will never change.

example 'a' value is 10 , whenever 'a' is presented corrsponding value will be10

Here some values missing in first column eg: NaN 10 which is a, NaN 40 which is d like wise dataframe contains 200 variables.

Values are not continuous variables, those are discrete and unsortable

In this case how can we impute missing values. Expected output should be:

Please help me on this.

Regards, Venkat.

Answer 1

I think in general it would be better to group and fill. We use DataFrame.groupby :

df.groupby('two').apply(lambda x: x.ffill().bfill())

---

It can be done without using groupby but you have to sort by both columns:

df.sort_values(['two','one']).ffill().sort_index()

---

Below I show you how the method proposed in another answer may fail:

Here is an example:

df=pd.DataFrame({'one':['a',np.nan,'c','d',np.nan,'c','b','b',np.nan,'a'],'two':[10,20,30,40,10,30,20,20,30,10]})
print(df)

   one  two
0    a   10
1  NaN   20
2    c   30
3    d   40
4  NaN   10
5    c   30
6    b   20
7    b   20
8  NaN   30
9    a   10

---

df.sort_values(['two']).fillna(method='ffill').sort_index()


  one  two
0   a   10
1   a   20
2   c   30
3   d   40
4   a   10
5   c   30
6   b   20
7   b   20
8   c   30
9   a   10

As you can see the proposed method in another of the answers fails here( see row 1 ). This occurs because some NaN Value can be the first for a specific value of the column 'two' and is filled with the value of the upper group.

This don't happen if we group first:

df.groupby('two').apply(lambda x: x.ffill().bfill())

  one  two
0   a   10
1   b   20
2   c   30
3   d   40
4   a   10
5   c   30
6   b   20
7   b   20
8   c   30
9   a   10

---

As I said we can use DataFrame.sort_values but we need to sort for both columns. I recommend you this method .

df.sort_values(['two','one']).ffill().sort_index()

  one  two
0   a   10
1   b   20
2   c   30
3   d   40
4   a   10
5   c   30
6   b   20
7   b   20
8   c   30
9   a   10

Answer 2

Here it is:

df.ffill(inplace=True)

output:

   one  two
0    a   10
1    b   20
2    c   30
3    d   40
4   aa   50
5   bb   60
6    a   10
7    b   20
8    c   30
9    d   40
10  aa   50

Answer 3

Try this:

df = df.sort_values(['two']).fillna(method='ffill').sort_index()

Which will give you

   one  two
0    a   10
1    b   20
2    c   30
3    d   40
4   aa   50
5   bb   60
6    a   10
7    b   20
8    c   30
9    d   40
10  aa   50