[英]How can I check a Pandas DataFrame's column against itself?
I have a Pandas DataFrame with two relevant columns. 我有一个带有两个相关列的Pandas DataFrame。 I need to check column A (a list of names) against itself, and if two (or more) values are similar enough to each other, I sum the values in column B for those rows. 我需要对照自身检查A列(名称列表),如果两个(或多个)值彼此足够相似,则将这些行的B列中的值求和。 To check similarity, I'm using the FuzzyWuzzy package that accepts two strings and returns a score. 为了检查相似性,我使用了FuzzyWuzzy包,该包接受两个字符串并返回一个分数。
Data: 数据:
a b
apple 3
orang 4
aple 1
orange 10
banana 5
I want to be left with: 我想留下:
a b
apple 4
orang 14
banana 5
I have tried the following line, but I keep getting a KeyError 我已经尝试了以下行,但是我一直收到KeyError
df['b']=df.apply(lambda x: df.loc[fuzz.ratio(df.a,x.a)>=70,'b'].sum(), axis=1)
I would also need to remove all rows where column b was added into another row. 我还需要删除将b列添加到另一行的所有行。
Any thoughts on how to accomplish this? 关于如何实现这一目标的任何想法?
Some parts here are best done with pandas, and some parts (eg, a function applied to a cartesian product) can be done without it. 这里的某些部分最好用熊猫来完成,而某些部分(例如,应用于笛卡尔积的函数)可以不用它来完成。
Overall, you can do this with: 总体而言,您可以使用以下方法执行此操作:
import itertools
import numpy as np
alias = {l : r for l, r in itertools.product(df.a, df.a) if l < r and
fuzz.ratio(l, r) > 70}
>>> df.b.groupby(df.a.replace(alias)).sum()
apple 4
banana 5
orange 14
Name: b, dtype: int64
The line 线
alias = {l : r for l, r in itertools.product(df.a, df.a) if l < r and
fuzz.ratio(l, r) > 70}
creates a map alias
, mapping words to their alias from a
. 创建地图alias
,映射字从他们别名a
。
The line 线
df.b.groupby(df.a.replace(alias)).sum()
groups b
by a translation using alias
, and then sums. 通过使用alias
的翻译将b
分组,然后求和。
I would map and groupby: 我会映射和分组:
def get_similarity(df, ind, col):
mapped = list(map(lambda x: fuzz.ratio(x, df[col].loc[ind]), df[col]))
cond = (np.array(mapped) >= 70)
label = df[col][cond].iloc[0]
return label
And use like this: 像这样使用:
df.groupby(lambda x: get_similarity(df, x, 'a'))['b'].sum()
声明:本站的技术帖子网页,遵循CC BY-SA 4.0协议,如果您需要转载,请注明本站网址或者原文地址。任何问题请咨询:yoyou2525@163.com.