GitHub

Getting Started

This guide walks through installing NiaARM.jl, loading data, and mining your first association rules.

Installation

Activate the project and install dependencies:

julia --project=. -e 'using Pkg; Pkg.instantiate()'

Or add the package directly in the REPL:

julia> using Pkg
julia> Pkg.add("NiaARM")

Loading data

NiaARM expects tidy tabular data. You can pass either a DataFrame or a path to a CSV file:

using NiaARM, CSV, DataFrames

transactions = CSV.read("datasets/sporty.csv", DataFrame)
dataset = Dataset(transactions)

The Dataset wrapper infers AbstractFeatures (numerical ranges or categorical values) and computes the problem dimensionality required by the optimizers.

Mining rules

  1. Choose an optimization algorithm (e.g. de, pso, ga, sa, randomsearch).
  2. Configure a StoppingCriterion (limit evaluations, iterations, or acceptable fitness).
  3. Pick metrics for fitness aggregation. Provide a vector of metric names (equal weights) or a Dict{Symbol,Float64} of weights.
using Random

criterion = StoppingCriterion(maxevals=5_000)
metrics = Dict(:support => 0.4, :confidence => 0.4, :lift => 0.2)

rules = mine(dataset, de, criterion; metrics=metrics, seed=1234)

for rule in rules[1:5]  # top 5
    println(rule, " | fitness=", rule.fitness, " support=", support(rule))
end

Picking metrics

Available metrics include support, confidence, coverage, lift, conviction, interestingness, yulesq, netconf, zhang, leverage, amplitude, inclusion, and comprehensibility. See Interestingness Measures for definitions and guidance.

Reusing discovered rules

Rules are returned as Rule objects with cached contingency tables. You can compute any metric later without re-evaluating transactions:

r = rules[1]
println(confidence(r))
println(rhs_support(r))

Next steps