arm-go

An implementation of FPGrowth frequent pattern generation algorithm, along with association rule generation, in Go.

Implementation is heavily optimized for performance.

This finds relationships of the form "people who buy X also buy Y", and also determines the strengths (confidence, lift, support) of those relationships.

For an overview of assocation rule mining, see Chapter 5 of Introduction to Data Mining, Kumar et al: Association Analysis: Basic Concepts and Algorithms.

The arm command line tool

To build, first you must download and install Go. Then install with:

go install github.com/cpearce/arm-go/cmd/arm

That will download, build, and install the arm binary into your $GOPATH/bin directory.

To run the binary, ensure $GOPATH/bin is in your path, and run:

arm \
  --input datasets/kosarak.csv \
  --output rules.csv \
  --itemsets itemsets.csv \
  --min-support 0.05 \
  --min-confidence 0.05 \
  --min-lift 1.5

Command line flags:

• input: path to CSV file containing transactions to analyze. There are some examples in the datasets/ directory.
• output: path to file to write the output rules to. Rules are written in CSV format with a header row explaining columns.
• itemsets: optional path to CSV file to write the generated frequent itemsets to. If specified the large itemsets are written to this file.
• min-support: minimum support above which itemsets are considered large, and used for rule generation.
• min-confidence: minimum confidence for rule generation.
• min-lift: minimum lift for rule generation.

The fpgrowth package

The underlying implementation can be used as a library as well. See the fpgrowth go package docs for more information.

Development

To run the cmd/arm binary:

go run github.com/cpearce/arm-go/cmd/arm \
  --input datasets/kosarak.csv \
  --output rules.csv \
  --itemsets itemsets.csv \
  --min-support 0.05 \
  --min-confidence 0.05 \
  --min-lift 1.5

Unit tests:

  $ go test ./...