Last updated 2014-06-05


Download #

Latest version available is Ver 2.0, released at 2014/06/05. Previous versions will not be needed as the current version covers all previous features, but should you need them, please contact us.
PREA gets ARFF format of input files. You can download sample datafiles in this section. Netflix dataset is a subset of the Netflix original dataset, taken from the first part. MovieLens dataset is exactly same with the original one. Note that time-dependent split files, using most recent ratings as test set while others as train set, are available only for Netflix data. For example, "netflix_3m1k_split_20050701.txt" uses ratings after 2005/07/01 as test set while all previous ones as train set. The idea is detailed in Gunawardana and Shani, JMLR 2009.

Environment #

This software is developed and tested in the environment described below. Without satisfying specification below or higher, we do not guarantee proper and/or efficient operation.
  • Intel(R) CoreTM2 Duo CPU E8400 @ 3.00GHz
  • 4.0GB of Main Memory

Operating System
  • Microsoft Windows 7 Service Pack 1 (32/64 bit)
  • Microsoft Windows XP Service Pack 3 (32/64 bit)
  • Ubuntu 10, with Linux 2.6.35-22 generic

Java Virtual Machine (JVM)
  • JavaTM 6 Standard Edition, Build 1.6.0_23

License #

PREA (Personalized Recommendation Algorithms Toolkit) is a free, open-source software, available under Free BSD License. You can modify and/or redistribute it without an explicit permission of the original authors. PREA can be distributed when it is expected to be useful, but without any warranty, merchantability, or fitness for a particular purpose. Please refer to full description of Free BSD License.
UJMP License
We use Universal Java Matrix Package (UJMP) for implementing dense matrix and its operations. UJMP is available under LGPL; details are available here: UJMP License page.