Package arules in R: getting rules with only one item in the left-hand side

Question

I am using the package arules in R to generate association rules. I would like to restrict the rules so that in the left-hand side there's only one particular element, let's call it "potatoe".

If I do this:

rules <- apriori(dtm.mat, parameter = list(sup = 0.4, conf =
0.9,target="rules"), appearance = list(lhs = c("potatoe")))

I get "potatoe" on the lhs, but also all other kinds of things. How can I force that the rules contain only one element? The parameter maxlen don't do what I want, because, as far as I can see, I cannot specify the maxlen to apply to the elements on the left.

Answer 1

Assuming that you've generated the rules ("rules", in your question), here is how to subset it. Basically, you have to coerce the data into a data frame and then subset it.

#Here are the original rules generated with some data I created
# categories are "G", "T", "D", and "potatoe"

> inspect(rules);
lhs        rhs      support    confidencelift
1  {}  => {T}       0.3333333  0.3333333 1.0000000
2  {}  => {G}       0.5000000  0.5000000 1.0000000
3  {}  => {potatoe} 0.5000000  0.5000000 1.0000000
4  {}  => {D}       0.5000000  0.5000000 1.0000000
5  {T} => {G}       0.1666667  0.5000000 1.0000000
6  {G} => {T}       0.1666667  0.3333333 1.0000000
7  {T} => {D}       0.1666667  0.5000000 1.0000000
8  {D} => {T}       0.1666667  0.3333333 1.0000000
9  {G} => {potatoe} 0.1666667  0.3333333 0.6666667
10 {potatoe} => {G} 0.1666667  0.3333333 0.6666667
11 {potatoe} => {D} 0.3333333  0.6666667 1.3333333
12 {D} => {potatoe} 0.3333333  0.6666667 1.3333333

#Coerce into data frame
as(rules, "data.frame");

#Restrict LHS to only certain value (here, "potatoe")
rules_subset <- subset(rules, (lhs %in% c("potatoe")));

#Check to see subset rules
inspect(rules_subset);
lhs            rhs support    confidencelift
1 {potatoe} => {G} 0.1666667  0.3333333 0.6666667
2 {potatoe} => {D} 0.3333333  0.6666667 1.3333333

This method also allows for arbitrarily many LHS values, not just one. Much easier than my previously proposed answer.