Preparing transaction data
In arules: Mining Association Rules and Frequent Itemsets

knitr::opts_chunk$set(collapse = TRUE, comment = "#>")
library(arules)
set.seed(1234)

Data for association rule mining comes from many sources and in several layouts. arules stores these data in the sparse transactions class, and the transactions() constructor accepts several common input layouts.

The following examples show how to convert each layout. Always inspect the resulting transactions object with summary() or itemLabels(): values that were encoded incorrectly in the source data may otherwise become unintended items.

A list of baskets

Use one character vector per transaction. List names become transaction IDs.

baskets <- list(
  order_1 = c("apple", "bread"),
  order_2 = c("bread", "milk"),
  order_3 = c("apple", "bread", "milk")
)
from_list <- transactions(baskets)
inspect(from_list)

Check both the transaction summary and the resulting item labels.

summary(from_list)
itemLabels(from_list)

The item labels confirm that the baskets were translated correctly.

A binary matrix

Rows represent transactions and columns represent items. Logical matrices make the intended coding explicit.

binary <- matrix(
  c(TRUE, TRUE, FALSE,
    FALSE, TRUE, TRUE,
    TRUE, TRUE, TRUE),
  nrow = 3,
  byrow = TRUE,
  dimnames = list(names(baskets), c("apple", "bread", "milk"))
)
from_matrix <- transactions(binary)

itemLabels(from_matrix)
inspect(from_matrix)

A data frame in wide format

Categorical columns are converted to items of the form variable=value. Logical columns represent the presence or absence of a single item. Missing values are omitted.

customers <- data.frame(
  age_group = factor(c("young", "adult", "adult")),
  region = factor(c("north", "south", "north")),
  subscriber = c(TRUE, FALSE, TRUE)
)
from_wide <- transactions(customers)

itemLabels(from_wide)
inspect(from_wide)

Continuous variables need to be discretized before conversion.

measurements <- data.frame(
  spend = c(12, 18, 35, 42, 55),
  visits = c(1, 2, 3, 5, 8)
)
measurements_discrete <- discretizeDF(
  measurements,
  default = list(method = "frequency", breaks = 2)
)
from_discrete <- transactions(measurements_discrete)

itemLabels(from_discrete)
inspect(from_discrete)

A data frame in long format

Long-format data has one row per transaction--item pair. Identify the transaction and item columns with cols.

long <- data.frame(
  order = c(1, 1, 2, 2, 3),
  product = c("apple", "bread", "bread", "milk", "apple")
)
from_long <- transactions(long, format = "long", cols = c("order", "product"))

itemLabels(from_long)
inspect(from_long)