Label encoding by value counts-CodePudding

I try to do label encoding for my cities. However, I want it to label according to which city is more than others. Let's say; Oslo has 500 rows Berlin has 400 rows Napoli has 300 rows in the dataset So label encoding will label those cities according to value counts so; Oslo should be labeled as 0, Berlin should be labeled 1, Napoli should labeled as 2

How I can do that?

CodePudding user response：

Use Series.map by Series with indices by Series.value_counts (sorted values by default):

df = pd.DataFrame({'col': ['Berlin'] * 4   ['Oslo'] * 5   ['Napoli'] * 3})
print (df)

s = df['col'].value_counts()
print (s)
Oslo      5
Berlin    4
Napoli    3
Name: col, dtype: int64

s1 = pd.Series(range(len(s)), index=s.index)
print (s1)
Oslo      0
Berlin    1
Napoli    2
dtype: int64
       
df['newcol'] = df['col'].map(s1)
print (df)
       col  newcol
0   Berlin       1
1   Berlin       1
2   Berlin       1
3   Berlin       1
4     Oslo       0
5     Oslo       0
6     Oslo       0
7     Oslo       0
8     Oslo       0
9   Napoli       2
10  Napoli       2
11  Napoli       2

Or use dictionary with enumerate:

s = df['col'].value_counts()
d = {v: k for k, v in enumerate(s.index)}
print (d)
{'Oslo': 0, 'Berlin': 1, 'Napoli': 2}      

df['newcol'] = df['col'].map(d)
print (df)
       col  newcol
0   Berlin       1
1   Berlin       1
2   Berlin       1
3   Berlin       1
4     Oslo       0
5     Oslo       0
6     Oslo       0
7     Oslo       0
8     Oslo       0
9   Napoli       2
10  Napoli       2
11  Napoli       2