Co-occurrence matrix using SAC?

Question

I have the following data frame 'x'

id,item,volume
a,c1,2
a,c2,3
a,c3,2
a,c4,1
a,c5,4
b,c6,6
b,c1,2
b,c3,1
b,c2,6
b,c4,4
c,c2,5
c,c8,6
c,c9,2
d,c1,1
e,c3,7
e,c2,3
e,c1,2
e,c9,5
e,c4,1
f,c1,7
f,c3,1

The first column is the id of a customer, the second column is the id of an item that customer bought and the third column is the number of those items bought. I'm trying to create a co-occurrence matrix which is a square matrix with 8 rows and columns, 8 being the number of distinct items.

n = length(unique(x$cid))

Could this be done through a SAC paradigm? For every id, I need to update the above matrix by adding +1 for every combination. For example for user 'b' with items c1,c2,c3,c4,c6, the 1st row in the matrix for columns 2,3,4 & 6 should be incremented by 1 and so on for all users. I'm unable to cast it into this framework. Any help much appreciated.

While R questions are perfectly welcome here, I wonder if this requires more statistics experience than programming experience; have you seen the [stats site](http://stats.stackexchange.com/faq)? Check the bit about picking which site to ask the question on .. if you think SO is fine, let me know and I'll delete my comment. If you think [stats.se] would be better, please flag your post for moderator attention and ask for a migration. — sarnold, May 13 '12 at 00:58
@sarnold I disagree, I think this is purely a question about rearranging data programmatically. I suspect the stats.SE folks would migrate it here. — joran, May 13 '12 at 01:25
To the OP, there are several aspects of your question that are unclear. You suggest you want the rows of your co-occurrence matrix to correspond to customers (id), but there are only 6 ids and you say you want an 8x8 matrix. Do you mean a 6x8 matrix? Also, should the entries of the matrix be 0/1, or the counts from the volume column? — joran, May 13 '12 at 01:27

user1317221_G · Accepted Answer · 2012-05-13T02:58:48.050

As @joran mentioned its kind of unclear what you really want. But If I understand correctly(probably not) I would like to propose an alternate route. I think you want a matrix of items to items with numbers representing the number of times the items share a customer?

If so then:

library(igraph)
library(tnet)
id_item<-cbind(
i=c(1,1,2,2,2,2,2,3,4,5,5,5,6,6,7,8),  #items
p=c(1,2,1,2,3,4,5,2,3,4,5,6,6,7,8,8))  #customers
item_item<-projecting_tm(id_item, method="sum")
item_item <- tnet_igraph(item_item,type="weighted one-mode tnet")
itemmat<-get.adjacency(item_item,attr="weight")
itemmat  #8x8 martrix of items to items

#     [,1] [,2] [,3] [,4] [,5] [,6] [,7] [,8]
#[1,]    0    2    1    0    0    0    0    0
#[2,]    2    0    1    1    2    0    0    0
#[3,]    1    1    0    0    0    0    0    0
#[4,]    0    1    0    0    0    0    0    0
#[5,]    0    2    0    0    0    1    0    0
#[6,]    0    0    0    0    1    0    0    0
#[7,]    0    0    0    0    0    0    0    1
#[8,]    0    0    0    0    0    0    1    0

If you need to take number of times the item was bought (volume) into account too then this is easily added into this code.

the final result is what I want. But can't get the packages to install on my version of ubuntu. Will poke around some more — broccoli, May 13 '12 at 16:33

Matthew Lundberg · Answer 2 · 2012-05-13T02:46:16.357

1

rehsape2 provides 'acast', which might be what you want:

> acast(x, id~item,value.var='volume')
  c1 c2 c3 c4 c5 c6 c8 c9
a  2  3  2  1  4 NA NA NA
b  2  6  1  4 NA  6 NA NA
c NA  5 NA NA NA NA  6  2
d  1 NA NA NA NA NA NA NA
e  2  3  7  1 NA NA NA  5
f  7 NA  1 NA NA NA NA NA

reshape2::dcast produces a data frame, if that is more convenient.

edited May 13 '12 at 02:46

answered May 13 '12 at 02:40

Matthew Lundberg

42,009
6
90
112

Co-occurrence matrix using SAC?

2 Answers2

Linked