使用模式填充缺失值 NAN 在 Pandas 中不起作用

Question

All the values is the df are one hot encoded, ie 0 / 1

Tried

fill_mode = lambda col: col.fillna(col.mode())
df = df.apply(fill_mode, axis=0)
df.isnull().sum()

Got

id      0
1           0
2           2
3           0

Expect all Null or NAN is filled with Mode.

Answer 1

col.mode() returns a series, not a single number. So col.fillna(col.mode()) will try to align the index of col.mode() with col and most likely you won't get anything updated. Maybe you want to do:

fill_mode = lambda col: col.fillna(col.mode()[0])

Answer 2

Adjust your fill_mode function

fill_mode = lambda col: col.fillna(col.mode().iloc[0])
df.apply(fill_mode, axis=0)

mode function return a series, fillna will match the index when received the series, however, in your case, we should remove the index match impacted.

Example

df=pd.DataFrame({'1':[np.nan,2,np.nan],'2':[1,1,np.nan]})

fill_mode = lambda col: col.fillna(col.mode())
print(df.apply(fill_mode, axis=0))
     1    2
0  2.0  1.0 # notice only the first item fill, since the out put of mode is index 0 with value 2
1  2.0  1.0
2  NaN  NaN 

df['1'].mode()
0    2.0
dtype: float64

In that case the df only fill the first value since the index matched.

We add the .iloc make it out put number, and will drop the index match with fillna

df['1'].mode().iloc[0]
2.0