在Python中使用模糊匹配合并多列数据框

时间:2019-01-05 08:22:53

标签: python pandas dataframe fuzzy-comparison

我有两个示例数据帧,如下所示:

df1 = pd.DataFrame({'Name': {0: 'John', 1: 'Bob', 2: 'Shiela'}, 
                   'Degree': {0: 'Masters', 1: 'Graduate', 2: 'Graduate'}, 
                   'Age': {0: 27, 1: 23, 2: 21}}) 

df2 = pd.DataFrame({'Name': {0: 'John S.', 1: 'Bob K.', 2: 'Frank'}, 
                   'Degree': {0: 'Master', 1: 'Graduated', 2: 'Graduated'}, 
                   'GPA': {0: 3, 1: 3.5, 2: 4}}) 

我想使用模糊匹配方法根据两列“名称”和“度”将它们合并在一起,以排除可能的重复项。这是我从这里的参考资料中得到的帮助: Apply fuzzy matching across a dataframe column and save results in a new column

from fuzzywuzzy import fuzz
from fuzzywuzzy import process

compare = pd.MultiIndex.from_product([df1['Name'],
                                      df2['Name']]).to_series()

def metrics(tup):
    return pd.Series([fuzz.ratio(*tup),
                      fuzz.token_sort_ratio(*tup)],
                     ['ratio', 'token'])
compare.apply(metrics)

compare.apply(metrics).unstack().idxmax().unstack(0)

compare.apply(metrics).unstack(0).idxmax().unstack(0)

让我们说一个人的名字和学位的fuzz.ratio都高于80,我们认为他们是同一个人。并将df1中的Name和Degree作为默认值。如何获得以下预期结果?谢谢。

df = df1.merge(df2, on = ['Name', 'Degree'], how = 'outer')

      Name     Degree   Age  GPA    duplicatedName   duplicatedDegree 
0     John    Masters  27.0  3.0         John S.          Master
1      Bob   Graduate  23.0  3.5          Bob K.         Graduated
2   Shiela   Graduate  21.0  NaN          NaN            Graduated
3    Frank  Graduated   NaN  4.0          NaN            Graduate

1 个答案:

答案 0 :(得分:2)

对于我工作60,我认为比率应该更低。用Series创建list comprehension,用N过滤并获得最大值。最后mapfillna和最后merge

from fuzzywuzzy import fuzz
from fuzzywuzzy import process
from  itertools import product

N = 60
names = {tup: fuzz.ratio(*tup) for tup in 
           product(df1['Name'].tolist(), df2['Name'].tolist())}

s1 = pd.Series(names)
s1 = s1[s1 > N]
s1 = s1[s1.groupby(level=0).idxmax()]

print (s1)
John S.    John
Bob K.      Bob
dtype: object

degrees = {tup: fuzz.ratio(*tup) for tup in 
           product(df1['Degree'].tolist(), df2['Degree'].tolist())}

s2 = pd.Series(degrees)
s2 = s2[s2 > N]
s2 = s2[s2.groupby(level=0).idxmax()]
print (s2)
Graduated    Graduate
Master        Masters
dtype: object

df2['Name'] = df2['Name'].map(s1).fillna(df2['Name'])
df2['Degree'] = df2['Degree'].map(s2).fillna(df2['Degree'])
#generally slowier alternative
#df2['Name'] = df2['Name'].replace(s1)
#df2['Degree'] = df2['Degree'].replace(s2)

df = df1.merge(df2, on = ['Name', 'Degree'], how = 'outer')
print (df)
     Name    Degree   Age  GPA
0    John   Masters  27.0  3.0
1     Bob  Graduate  23.0  3.5
2  Shiela  Graduate  21.0  NaN
3   Frank  Graduate   NaN  4.0