Pandas: How to convert a one-hot-encoded dataframe to an adjacency matrix?

Question

I have a data frame df, of the form:

      Col1   Col2    Col3   
0      0       1       0
1      1       1       0
2      0       1       1
3      1       1       0

I need a new df of the form:

        Col1    Col2    Col3
Col1      0       2       0    
Col2      2       0       1
Col3      0       1       0

Basically the values represent the co-occurrences of two given columns for all rows.

How do I go about this?

Why is `Col2/Col1` equal to 2? – Oliver Charlesworth Dec 22 '17 at 00:27 — Oliver Charlesworth, Dec 22 '17 at 00:27
Sorry, I had missed a row. Fixed now. – Melsauce Dec 22 '17 at 00:38 — Melsauce, Dec 22 '17 at 00:38

score 4 · Accepted Answer · answered Dec 22 '17 at 06:38

4

Simply leverage matrix-multiplication there -

In [21]: df_out = df.T.dot(df)

In [22]: np.fill_diagonal(df_out.values, 0)

In [23]: df_out
Out[23]: 
      Col1  Col2  Col3
Col1     0     2     0
Col2     2     0     1
Col3     0     1     0

answered Dec 22 '17 at 06:38

Divakar

218,885
19
262
358

1

Beat me to it ! – Paul Panzer Dec 22 '17 at 06:41
Great, I try to using dot, but forget the T :-) – BENY Dec 22 '17 at 15:18
@Wen Yeah I have seen you use that and was surprised to not see this time around from you. – Divakar Dec 22 '17 at 15:19
What I have try to do is `df.dot(df.columns.values)`, but this is really hard to split . :-) – BENY Dec 22 '17 at 15:22

Pandas: How to convert a one-hot-encoded dataframe to an adjacency matrix?

1 Answers1