我目前有2个数据帧,1个用于捐赠者,1个用于筹款。理想情况下,我想要找到的是,如果有任何筹款人也捐赠,如果是的话,将一些信息复制到我的募捐人数据集(捐赠者姓名,电子邮件和他们的第一次捐赠)。我的数据存在问题1)我需要通过姓名和电子邮件进行匹配,但用户可能会略有不同的名称(来自Kat和Kathy)。 2)捐赠者和筹款人的名称重复。 2a)有了捐赠者,我可以得到唯一的姓名/电子邮件组合,因为我只关心第一个捐赠日期2b)虽然我需要保留两行而不会丢失数据,如日期。
我现在的示例代码:
import pandas as pd
import datetime
from fuzzywuzzy import fuzz
import difflib
donors = pd.DataFrame({"name": pd.Series(["John Doe","John Doe","Tom Smith","Jane Doe","Jane Doe","Kat test"]), "Email": pd.Series(['[email protected]','[email protected]','[email protected]','[email protected]','[email protected]','[email protected]']),"Date": (["27/03/2013 10:00:00 AM","1/03/2013 10:39:00 AM","2/03/2013 10:39:00 AM","3/03/2013 10:39:00 AM","4/03/2013 10:39:00 AM","27/03/2013 10:39:00 AM"])})
fundraisers = pd.DataFrame({"name": pd.Series(["John Doe","John Doe","Kathy test","Tes Ester", "Jane Doe"]),"Email": pd.Series(['[email protected]','[email protected]','[email protected]','[email protected]','[email protected]']),"Date": pd.Series(["2/03/2013 10:39:00 AM","27/03/2013 11:39:00 AM","3/03/2013 10:39:00 AM","4/03/2013 10:40:00 AM","27/03/2013 10:39:00 AM"])})
donors["Date"] = pd.to_datetime(donors["Date"], dayfirst=True)
fundraisers["Date"] = pd.to_datetime(donors["Date"], dayfirst=True)
donors["code"] = donors.apply(lambda row: str(row['name'])+' '+str(row['Email']), axis=1)
idx = donors.groupby('code')["Date"].transform(min) == donors['Date']
donors = donors[idx].reset_index().drop('index',1)
因此,这给了我每个捐赠者的第一次捐赠(假设任何具有完全相同名称和电子邮件的人都是同一个人)。
理想情况下,我希望我的筹款人数据集看起来像:
Date Email name Donor Name Donor Email Donor Date
2013-03-27 10:00:00 [email protected] John Doe John Doe [email protected] 2013-03-27 10:00:00
2013-01-03 10:39:00 [email protected] John Doe John Doe [email protected] 2013-03-27 10:00:00
2013-02-03 10:39:00 [email protected] Kathy test Kat test [email protected] 2013-03-27 10:39:00
2013-03-03 10:39:00 [email protected] Tes Ester
2013-04-03 10:39:00 [email protected] Jane Doe Jane Doe [email protected] 2013-04-03 10:39:00
我试着跟随这个帖子:is it possible to do fuzzy match merge with python pandas?但是不断让索引超出范围错误(猜测它不喜欢筹款活动中的重复名称):(那么任何想法如何匹配/合并这些数据集?
用for循环做它(它工作但速度很慢,我觉得必须有更好的方法)
fundraisers["donor name"] = ""
fundraisers["donor email"] = ""
fundraisers["donor date"] = ""
for donindex in range(len(donors.index)):
max = 75
for funindex in range(len(fundraisers.index)):
aname = donors["name"][donindex]
comp = fundraisers["name"][funindex]
ratio = fuzz.ratio(aname, comp)
if ratio > max:
if (donors["Email"][donindex] == fundraisers["Email"][funindex]):
ratio *= 2
max = ratio
fundraisers["donor name"][funindex] = aname
fundraisers["donor email"][funindex] = donors["Email"][donindex]
fundraisers["donor date"][funindex] = donors["Date"][donindex]
这里有一些pythonic(在我看来),工作(在你的例子中)代码,没有显式循环:
def get_donors(row):
d = donors.apply(lambda x: fuzz.ratio(x['name'], row['name']) * 2 if row['Email'] == x['Email'] else 1, axis=1)
d = d[d >= 75]
if len(d) == 0:
v = ['']*3
else:
v = donors.ix[d.idxmax(), ['name','Email','Date']].values
return pd.Series(v, index=['donor name', 'donor email', 'donor date'])
pd.concat((fundraisers, fundraisers.apply(get_donors, axis=1)), axis=1)
输出:
Date Email name donor name donor email donor date
0 2013-03-27 10:00:00 [email protected] John Doe John Doe [email protected] 2013-03-01 10:39:00
1 2013-03-01 10:39:00 [email protected] John Doe John Doe [email protected] 2013-03-01 10:39:00
2 2013-03-02 10:39:00 [email protected] Kathy test Kat test [email protected] 2013-03-27 10:39:00
3 2013-03-03 10:39:00 [email protected] Tes Ester
4 2013-03-04 10:39:00 [email protected] Jane Doe Jane Doe [email protected] 2013-03-04 10:39:00
我会使用Jaro-Winkler,因为它是目前最常用且最准确的近似字符串匹配算法之一[Cohen, et al.],[Winkler]。
这是我用jellyfish包中的Jaro-Winkler来做的:
def get_closest_match(x, list_strings):
best_match = None
highest_jw = 0
for current_string in list_strings:
current_score = jellyfish.jaro_winkler(x, current_string)
if(current_score > highest_jw):
highest_jw = current_score
best_match = current_string
return best_match
df1 = pandas.DataFrame([[1],[2],[3],[4],[5]], index=['one','two','three','four','five'], columns=['number'])
df2 = pandas.DataFrame([['a'],['b'],['c'],['d'],['e']], index=['one','too','three','fours','five'], columns=['letter'])
df2.index = df2.index.map(lambda x: get_closest_match(x, df1.index))
df1.join(df2)
输出:
number letter
one 1 a
two 2 b
three 3 c
four 4 d
five 5 e
更新:使用Levenshtein模块中的jaro_winkler来提高性能。
from jellyfish import jaro_winkler as jf_jw
from Levenshtein import jaro_winkler as lv_jw
%timeit jf_jw("appel", "apple")
>> 339 ns ± 1.04 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
%timeit lv_jw("appel", "apple")
>> 193 ns ± 0.675 ns per loop (mean ± std. dev. of 7 runs, 1000000 loops each)
如何使用Pandas识别DataFrame中的模糊复制
def get_ratio(row):
name = row['Name_1']
return fuzz.token_sort_ratio(name,"Ceylon Hotels Corporation")
df[df.apply(get_ratio, axis=1) > 70]