In arules how to turn a sparse dataframe into transactions?

0
votes

Hi I have a sparse dataframe of grocery order like this

library(arules)
a_df <- data.frame(
apple = as.factor(c(1,0,0,0,1,1)),
banana = as.factor(c(0,1,1,0,0,0)),
peeler = as.factor(c(1,0,0,0,1,1)))

a_tran = as(a_df, "transactions" )
inspect(a_tran)
rules <- apriori(a_tran, parameter=list(minlen=2, supp=0.5,conf = 0.5))
inspect(rules)

However the result include 0s (the item not ordered) like this: lhs rhs support confidence lift count {banana=0} => {apple=1} 0.5 0.6 1.2 3

How can I ignore the 0s in the dataframe, or transform the dataframe to something like 

order 1: apple, peeler
order 2: banana

Thanks.

2
rarulesmarket-basket-analysis

2 Answers

0
votes

Here are a few options

library(magrittr)
idx <- which(a_df==1, arr.ind = T)
(lst <- split(names(a_df)[idx[,2]], idx[,1]))
# $`1`
# [1] "apple"  "peeler"
# 
# $`2`
# [1] "banana"
# 
# $`3`
# [1] "banana"
# 
# $`5`
# [1] "apple"  "peeler"
# 
# $`6`
# [1] "apple"  "peeler"

rules <- function(x, app=NULL) { 
  x %>% as("transactions") %>% apriori(parameter=list(minlen=2, supp=0.5,conf = 0.5), appearance=app) 
}
# use a list without "0"s:
lst %>% rules %>% inspect
# filter "0"s afterwards:
a_df %>% rules %>% subset(!lhs%pin%"0" & !rhs%pin%"0") %>% inspect
# filter "0"s in apriori:
a_df %>% rules(list(none = paste(names(a_df), "0", sep="="), default="both")) %>% inspect
1
votes

Looks like your data is a full 0-1 matrix. Here is the fastest way:

trans <- as(a_df == "1", "transactions")
inspect(trans)

    items         
[1] {apple,peeler}
[2] {banana}      
[3] {banana}      
[4] {}            
[5] {apple,peeler}
[6] {apple,peeler}

Now you can mine rules.