为从 pandas 开始的无向图创建节点

Question

I have a dataframe that looks like this (I have 170000 observations in reality):

Firm   pat             cited_pat
F_1    [p0,p1,p2]       [p0,p1,p2]    
F_2    []               []
F_3    [p3,p6,p2]       [p5,p0,p23,p29,p12,p8]
F_4    [p0,p9,p25]      [p0,p29,p31]
...

The idea is this:

1. Create all possible couples of F_i, F_j;
2. If two F_i, F_j have one (or more) "ps" in common, then put an edge of 1 and stop;
3. If they do not, then take cited_pat and check how many "ps" are in common there. If more than 50% are in common than create an edge=1.

Now, I am struggling a lot finding aa way to do it in an easy way. Could you please help me on this?

Answer 1

Here's one way to do things:

import pandas as pd
import numpy as np
import networkx as nx

data = {'Firm': {0: 'F_1', 1: 'F_2', 2: 'F_3', 3: 'F_4'},
 'pat': {0: ['p0','p1','p2'], 1: [], 2: ['p3','p6','p2'], 3: ['p0','p9','p25']},
 'cited_pat': {0: ['p0','p1','p2'],
  1: [],
  2: ['p5','p0','p23','p29','p12','p8'],
  3: ['p0','p29','p31']}}

df = pd.DataFrame(data)

def cited_pat_func(set_i):
    def f(set_j):
        return len(set_i & set_j)*2 >= len(set_i | set_j)
    return f

G = nx.Graph()
G.add_nodes_from(df['Firm'])

for i,row in df.iterrows():
    df_tail = df.iloc[(i+1):,:]
    F_i = row['Firm']
    pat_i = set(row['pat'])
    cpat_i = set(row['cited_pat'])
    
    cond = (df_tail['pat'].apply(set)
              .apply(pat_i.intersection)
              .astype(bool) |
            df_tail['cited_pat'].apply(set)
              .apply(cited_pat_func(cpat_i)))
    for F_j in df_tail.loc[cond,'Firm']:
        G.add_edge(F_i, F_j)

Here's the graph produced for this example: