jLCM

A Java implementation of the LCM (Linear Closed itemsets Miner) algorithm, as proposed by T.Uno & H.Arimura. It is multi-threaded, as proposed by Négrevergne et al., hence the name of its main class: PLCM.

Key features:

• Mines closed frequent itemsets
• Excellent speed-up on multi-core machines
• Provides iterator interfaces to adapt your own items/transactions/itemset collections
• Able to load datasets greater than 4GB (since v1.7)

Reference papers:

• "An efficient algorithm for enumerating closed patterns in transaction databases" by T. Uno, T. Asai, Y. Uchida and H. Arimura, in Discovery Science, 2004
• "Discovering closed frequent itemsets on multicore: Parallelizing computations and optimizing memory accesses" by B. Negrevergne, A. Termier, J-F. Mehaut, and T. Uno in International Conference on High Performance Computing & Simulation, 2010

Please use Maven to build the program.

jLCM as a command-line utility

Download jLCM-cli's JAR and invoke java -jar jLCM-cli-1.7.0-wdeps.jar to show the complete manual. Note that this program's main function lives in a separated project.

This tool uses ASCII files as input: each line represents a transaction (using UNIX line terminators). You may find example input files in the FIMI repository, or start with a small one embedded in src/test/resources like 50retail.dat.

jLCM as a library/Maven dependency

Add the following dependency to your pom.xml

<dependency>
    <groupId>fr.liglab.jlcm</groupId>
    <artifactId>jLCM</artifactId>
    <version>1.7.0</version>
</dependency>

To perform the mining you will have to instanciate an ExplorationStep, a PatternsCollector and the main class PLCM. Depending on how you want to do the I/O you may have to implement your own Iterable<TransactionReader> (for input) and/or PatternsWriter (for output).

The main class of jLCM-cli provides an example use of the library.

License and copyright owners

This work is released under the Apache License 2.0 (see LICENSE).

Copyright 2013,2014,2015,2016 Martin Kirchgessner, Vincent Leroy, Alexandre Termier, Sihem Amer-Yahia, Marie-Christine Rousset, Université Joseph Fourier and CNRS.