Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

How can I convert a two column array to a matrix with counts of occurences?

I have the following numpy array:

import numpy as np

pair_array = np.array([(205, 254), (205, 382), (254, 382), (18, 69), (205, 382), 
                       (31, 183), (31, 267), (31, 382), (183, 267), (183, 382)])

print(pair_array)

#[[205 254]
# [205 382]
# [254 382]
# [ 18  69]
# [205 382]
# [ 31 183]
# [ 31 267]
# [ 31 382]
# [183 267]
# [183 382]]

Is there a way to transform this array to a symmetric pandas Dataframe that contains the count of occurences for all possible combinations? I expect something along the lines of this:

#     18  31  69 183 205 254 267 382 
#  18  0   0   1   0   0   0   0   0
#  31  0   0   0   1   0   0   1   1
#  69  1   0   0   0   0   0   0   0
# 183  0   1   0   0   0   0   1   1
# 205  0   0   0   0   0   1   0   2
# 254  0   0   0   0   1   0   0   1
# 267  0   1   0   1   0   0   0   0
# 382  0   1   0   1   2   1   0   0
like image 534
M. Saïd H. Unger Avatar asked Oct 21 '20 13:10

M. Saïd H. Unger


4 Answers

One way could be to build a graph using NetworkX and obtain the adjacency matrix directly as a dataframe with nx.to_pandas_adjacency. To account for the co-occurrences of the edges in the graph, we can create a nx.MultiGraph, which allows for multiple edges connecting the same pair of nodes:

import networkx as nx

G = nx.from_edgelist(pair_array, create_using=nx.MultiGraph)
nx.to_pandas_adjacency(G, nodelist=sorted(G.nodes()), dtype='int')

      18   31   69   183  205  254  267  382
18     0    0    1    0    0    0    0    0
31     0    0    0    1    0    0    1    1
69     1    0    0    0    0    0    0    0
183    0    1    0    0    0    0    1    1
205    0    0    0    0    0    1    0    2
254    0    0    0    0    1    0    0    1
267    0    1    0    1    0    0    0    0
382    0    1    0    1    2    1    0    0

Building a NetworkX graph, will also enable to create an adjacency matrix or another depending on the behaviour we expect. We can either create it using a:

  • nx.Graph: If we want to set to 1 both entries (x,y) and (y,x) for a (x,y) (or (y,x)) edge. This will hence produce a symmetric adjacency matrix
  • nx.DiGraph: If (x,y) should only set the (x,y) the entry to 1
  • nx.MultiGraph: For the same behaviour as a nx.Graph but accounting for edge co-occurrences
  • nx.MultiDiGraph: For the same behaviour as a nx.DiGraph but also accounting for edge co-occurrences
like image 110
yatu Avatar answered Oct 20 '22 00:10

yatu


One way of doing it is appending the pair_array with pair_array reversed at axis 1 which can be done using [::-1]. And to append use np.vstack/np.r_/np.concatenate.

Now use pd.crosstab to perform cross tabulation.

all_vals = np.r_[pair_array, pair_array[:, ::-1]]
pd.crosstab(all_vals[:, 0], all_vals[:, 1])

col_0  18   31   69   183  205  254  267  382
row_0                                        
18       0    0    1    0    0    0    0    0
31       0    0    0    1    0    0    1    1
69       1    0    0    0    0    0    0    0
183      0    1    0    0    0    0    1    1
205      0    0    0    0    0    1    0    2
254      0    0    0    0    1    0    0    1
267      0    1    0    1    0    0    0    0
382      0    1    0    1    2    1    0    0

As @QuangHoang pointed when there are identical pairs occurring more than one time i.e [(18, 18), (18, 18), ...], then use

rev = pair_array[:, ::-1]
m = (pair_array == rev)
rev = rev[~np.all(m, axis=1)]
all_vals = np.r_[pair_arr, rev]
like image 31
Ch3steR Avatar answered Oct 19 '22 22:10

Ch3steR


You could create a data frame of the appropriate size with zeros beforehand and just increment the appropriate cells by looping over the pairs:

import numpy as np
import pandas as pd

pair_array = np.array([(205, 254), (205, 382), (254, 382), (18, 69), (205, 382),
                       (31, 183), (31, 267), (31, 82), (183, 267), (183, 382)])

vals = sorted(set(pair_array.flatten()))
n = len(vals)

df = pd.DataFrame(np.zeros((n, n), dtype=np.int), columns=vals, index=vals)

for r, c in pair_array:
    df.at[r, c] += 1
    df.at[c, r] += 1

print(df)

Output:

     18   31   69   82   183  205  254  267  382
18     0    0    1    0    0    0    0    0    0
31     0    0    0    1    1    0    0    1    0
69     1    0    0    0    0    0    0    0    0
82     0    1    0    0    0    0    0    0    0
183    0    1    0    0    0    0    0    1    1
205    0    0    0    0    0    0    1    0    2
254    0    0    0    0    0    1    0    0    1
267    0    1    0    0    1    0    0    0    0
382    0    0    0    0    1    2    1    0    0
like image 2
Sash Sinha Avatar answered Oct 20 '22 00:10

Sash Sinha


This is crosstab:

pd.crosstab(pair_array[:,0], pair_array[:,1])

Output:

col_0  69   82   183  254  267  382
row_0                              
18       1    0    0    0    0    0
31       0    1    1    0    1    0
183      0    0    0    0    1    1
205      0    0    0    1    0    2
254      0    0    0    0    0    1
like image 1
Quang Hoang Avatar answered Oct 19 '22 23:10

Quang Hoang