ARules

1. Installation

julia> Pkg.add("https://github.com/bcbi/ARules.jl")

2. Frequent Itemset Generation

The frequent() function can be used to obtain frequent itemsets using the a priori algorithm. The second and third arguments allow us to control the minimum support threshold (either as a count or proportion) and the maximum size of itemset to consider, respectively.

julia> using ARules

julia> transactions = [["milk", "eggs", "bread"],
                       ["butter", "milk", "sugar", "flour", "eggs"],
                       ["bacon", "eggs", "milk", "beer"],
                       ["bread", "ham", "turkey"],
                       ["cheese", "ham", "bread", "ketchup"],
                       ["mustard", "hot dogs", "buns", "hamburger", "cheese", "beer"],
                       ["milk", "sugar", "eggs"],
                       ["hamburger", "ketchup", "milk", "beer"],
                       ["ham", "cheese", "bacon", "eggs"]]

julia> frequent(transactions, 2, 6)             # uses a-priori algorithm

3. Association Rule Generation

The apriori() function can be used to obtain association rules.

julia> using ARules

julia> transactions = [["milk", "eggs", "bread"],
                       ["butter", "milk", "sugar", "flour", "eggs"],
                       ["bacon", "eggs", "milk", "beer"],
                       ["bread", "ham", "turkey"],
                       ["cheese", "ham", "bread", "ketchup"],
                       ["mustard", "hot dogs", "buns", "hamburger", "cheese", "beer"],
                       ["milk", "sugar", "eggs"],
                       ["hamburger", "ketchup", "milk", "beer"],
                       ["ham", "cheese", "bacon", "eggs"]]

julia> rules = apriori(transactions, supp = 0.01, conf = 0.1, maxlen = 6)

4. Note

This package is under active development. And as such, there are still many performance and feature improvements to be made. In the case of performance, while the package will handle many applications quite well, once the number of "items" in "transactions" becomes large, there is a marked performance penalty.

5. To Do

• Implement additional frequent-itemset generation algorithms (e.g., eclat, fp-growth)
• Add functionality for requiring rules to contain a certain item (or items)
• Improve performance