在 itertools.combinations Python 中求和值

Question

我有这个数据集：

import numpy as np
import pandas as pd
from itertools import product

A= ['ABC', 'DEF'] 
M= ['X', 'Y', 'Z']
F= ['plus', 'minus', 'star']

# Create all possible permutation from <A,M,F> 
df = pd.DataFrame(list(product(A,M,F)), columns=['A', 'M', 'F'])
df['value'] = np.random.uniform(0, 1, df.shape[0])

数据集如下：

     A  M   F        value
0   ABC X   plus    0.666602
1   ABC X   minus   0.716765
2   ABC X   star    0.032931
3   ABC Y   plus    0.275616
4   ABC Y   minus   0.489233

在这里，我想获得能够最大化我的目标的前 k 个集合组合：

My goal is : The maximum of Sum(values of combination sets) + sum(distance of combination sets)

这是我的代码：

#diversity/distance function
def diversity(a, b):
    c = a.intersection(b)
    d = float(len(c)) / (len(a) + len(b) - len(c))
    return 1 - d

我的代码：

from itertools import combinations

k = 3

max_distance = []

# I drop the column 'value' because sets that I want to compare is <A,M,F>
df_distance = df.drop(['value'],axis=1)
series_set = df_distance.apply(lambda row: set(row), axis=1)
data = series_set

for z in combinations(data, k):
    dis = 0
    sum_values = 0
    for a in combinations(z, 2):
        dis += diversity(*a)
        # I am stuck here, I want to sum the value but I don't know, how to get the value and sum it in combination
    max_distance.append((dis, tuple(z)))

max_distance.sort(key=lambda x: x[0], reverse=True)
print(max_distance[:k])

输出：

[(2.8, ({'plus', 'ABC', 'X'}, {'Y', 'minus', 'ABC'}, {'Z', 'star', 'DEF'})), (2.8, ({'plus', 'ABC', 'X'}, {'Y', 'star', 'ABC'}, {'Z', 'minus', 'DEF'})), (2.8, ({'plus', 'ABC', 'X'}, {'Z', 'minus', 'ABC'}, {'Y', 'star', 'DEF'}))]

在我上面的代码中，我只是计算距离的总和。值 2.8 只是距离的总和。我想对集合之间的距离求和，但只能从列 [A，M，F] 中求和，我还想对这些值求和。预期输出是（距离之和 + 值之和），它是所有集合组合的最佳值。

我真的很困惑如何对组合中的值求和？

预期输出：

  [(sum(distance) + sum(values) , ({'plus', 'ABC', 'X'}, {'Y', 'minus', 'ABC'}, {'Z', 'star', 'DEF'})), ((sum(distance) + sum(values), ({'plus', 'ABC', 'X'}, {'Y', 'star', 'ABC'}, {'Z', 'minus', 'DEF'})), ((sum(distance) + sum(values), ({'plus', 'ABC', 'X'}, {'Z', 'minus', 'ABC'}, {'Y', 'star', 'DEF'}))]

如果您有任何问题，请告诉我，对不起我的英语。

Answer 1

请参阅下面稍微修改过的代码版本。我认为这是你想要的。我基本上将您的 set 转换为多样性函数，以便 series_set 可以是一个元组。然后可以使用该元组对具有多索引的 DataFrame 进行切片。

import numpy as np
import pandas as pd
from itertools import product, combinations

A = ['ABC', 'DEF']
M = ['X', 'Y', 'Z']
F = ['plus', 'minus', 'star']

# Create all possible permutation from <A,M,F>
df = pd.DataFrame(list(product(A,M,F)), columns=['A', 'M', 'F'])
df['value'] = np.random.uniform(0, 1, df.shape[0])


# diversity/distance function
def diversity(a, b):
    c = set(a).intersection(b)
    d = float(len(c)) / (len(a) + len(b) - len(c))
    return 1 - d

k = 3
max_distance = []
max_values = []

# I drop the column 'value' because sets that I want to compare is <A,M,F>
df_distance = df.drop(['value'],axis=1)
df_sum = df.set_index(['A', 'M', 'F'])
series_set = df_distance.apply(lambda row: tuple(row), axis=1)
data = series_set

for z in combinations(data, k):
    dis = 0
    sum_values = 0
    for a in combinations(z, 2):
        dis += diversity(*a)
        sum_values += df_sum.ix[a[0], 'value'] + df_sum.ix[a[1], 'value']
    max_distance.append((dis, tuple(z)))
    max_values.append((sum_values, tuple(z)))

max_distance.sort(key=lambda x: x[0], reverse=True)
print(max_distance[:k])

max_values.sort(key=lambda x: x[0], reverse=True)
print(max_values[:k])

-- 更新--

max_total = []
for z in combinations(data, k):
    dis = 0
    sum_values = 0
    for a in combinations(z, 2):
        dis += diversity(*a)
        sum_values += df_sum.loc[a[0], 'value'] + df_sum.loc[a[1], 'value']
    total_sum = dis + sum_values
    max_total.append((total_sum, tuple(z)))

max_total.sort(key=lambda x: x[0], reverse=True)
print(max_total[:k])