How to calculate the Jaccard Index in Python?

Question

I have a data set as I've shown below:

It shows which book is sold by which shop.

import pandas as pd

books = {'shop': ["A", "B", "C", "D", "E", "A", "B", "C", "D",],
        'book_id': [1, 1, 2, 3, 3, 3, 4, 5, 1,]
        }

df = pd.DataFrame(books, columns = ['shop', 'book_id'])

Here is the print:

  shop  book_id
0    A        1
1    B        1
2    C        2
3    D        3
4    E        3
5    A        3
6    B        4
7    C        5
8    D        1

In the data set,

shop A sells 1, 3
shop B sells 1, 4
shop C sells 2, 5
shop D sells 3, 1
shop E sells only 3

So now, I want to calculate the jaccard index here. For instance, let's take shop A and shop B. There are three different books that are sold by A and B (book 1, book 3, book 4). However, only one product is sold by both shops (this is product 1). So, the Jaccard index here should be 33.3% (1/3).

Here is the sample of the desired data:

result = {'shop_1': ["A", "B", "A", "C", "A", "D", "A", "E",],
          'shop_2': ["B", "A", "C", "A", "D", "A", "E", "A",],
          'jaccard':  [33.3, 33.33, 0, 0, 100, 100, 50, 50,]
        }
desired_df = pd.DataFrame(result, columns = ['shop_1', 'shop_2', 'jaccard'])

Print
  shop_1 shop_2  jaccard
0      A      B    33.30
1      B      A    33.33
2      A      C     0.00
3      C      A     0.00
4      A      D   100.00
5      D      A   100.00
6      A      E    50.00
7      E      A    50.00
.      .      .      .
.      .      .      .
.      .      .      .

Can someone help me to do this? Is there any library to implement Jaccard Index?

user_4238408 · Answer

If you data is not too big, you can use a broadcasting approach:

books = pd.crosstab(df.shop, df.book_id)

# underlying numpy
arr = books.values

common = (arr[None,...] | arr[:,None,:]).sum(-1)

output = (books @ books.T)/common

Output:

shop         A         B    C         D    E
shop                                        
A     1.000000  0.333333  0.0  1.000000  0.5
B     0.333333  1.000000  0.0  0.333333  0.0
C     0.000000  0.000000  1.0  0.000000  0.0
D     1.000000  0.333333  0.0  1.000000  0.5
E     0.500000  0.000000  0.0  0.500000  1.0

To match your expected output:

output = (output.stack().rename_axis(['shop_1','shop_2'])
                .reset_index(name='jaccard')
                .query('shop_1 != shop_2')
         )

Output:

   shop_1 shop_2   jaccard
1       A      B  0.333333
2       A      C  0.000000
3       A      D  1.000000
4       A      E  0.500000
5       B      A  0.333333
7       B      C  0.000000
8       B      D  0.333333
9       B      E  0.000000
10      C      A  0.000000
11      C      B  0.000000
13      C      D  0.000000
14      C      E  0.000000
15      D      A  1.000000
16      D      B  0.333333
17      D      C  0.000000
19      D      E  0.500000
20      E      A  0.500000
21      E      B  0.000000
22      E      C  0.000000
23      E      D  0.500000

How to calculate the Jaccard Index in Python?

1 Answers