Pfthb

Consider this simple example

pd.DataFrame('id' : [1,1,2,3,4],
 'place' : ['bar','pool','bar','kitchen','bar'])

Out[4]: 
 id place
0 1 bar
1 1 pool
2 2 bar
3 3 kitchen
4 4 bar

Here the network structure is such that a given id is connected to another id if they went to the same place.

For instance, here 1 is connected to 2 and 4 because they are at the bar.

1 and 3 are NOT connected because 1 went to bar and pool which does not include kitchen (the only place 3 went to)

My real data is huge, about 500k. What is the most efficient way to proceed to get the adjacency list? Here this is just a string with the format source target target like in https://networkx.github.io/documentation/networkx-1.10/reference/readwrite.adjlist.html

adjacency_list
1 2 4
2 1 4
4 1 2

Can we avoid loops and use Pandas tricks?

Thanks!

Using unique then switch the column 0 to 1 and column 1 to 0 concat the both df together

adj=pd.DataFrame(df.groupby('place').id.unique().loc[lambda x : x.str.len()>1].tolist())
pd.concat([adj,adj.rename(columns=0:1,1:0)])
Out[810]: 
 0 1
0 1 2
0 2 1

Update :

newdf=df.merge(df,on='place')
x=nx.from_pandas_dataframe(newdf,'id_x','id_y') # using merge to get the connect for all id by link columns place. 
[list(itertools.permutations(x, len(x)) for x in list(nx.connected_components(x))] # using permutations get the all combination for each connected_components in networkx 
Out[821]: [[(1, 2), (1, 3), (2, 1), (2, 3), (3, 1), (3, 2)]]

Data input

df
Out[822]: 
 id place
0 1 bar
1 1 pool
2 2 bar
3 3 bar

What about:

>>> df
 id place
0 1 bar
1 1 pool
2 2 bar
3 3 kitchen
>>> df.groupby('place').id.nunique().value_counts()
1 2
2 1
Name: id, dtype: int64