PRA (Path Ranking Algorithm)
An implementation of the Path Ranking Algorithm (PRA) using GraphChi, a library for efficient processing of large graphs on a single (large) machine. This algorithm learns models that analyze a graph and predict missing edges of particular types. This codebase also implements Subgraph Feature Extraction (SFE), a method similar to PRA that is simpler and faster.
See the navigation on the left for code documentation. Please feel free to file bugs, feature requests, or send pull requests. And, if you find some of this documentation confusing or lacking, please feel free to contact me. I try to be responsive to reasonable requests for help, and I would like for this documentation to be useful; if it's not, tell me.
NOTE, however: the code tends to change frequently, and it's generally more work than it's worth to
keep the documentation up to date, because it gets little use. I mostly just update it when
someone asks me questions. At the moment, you're probably better off starting from the quick
start instructions for reproducing my 2015 EMNLP paper, then
checking here (or in the
examples/ directory in the code) to see what parameters are available if
you want to change things.
Using the code
If all you want to do is run this algorithm as-is on your dataset, see the getting started
guide. If you want to use this code as a library in your
own system, you can easily do so by using maven or
sbt as your build system. The code is released on the maven central
repository, so all you have to do is include it as a managed dependency. If you are using sbt,
include the following line in your
libraryDependencies += "edu.cmu.ml.rtw" %% "pra" % "3.3"
If you are using maven, you can include the PRA code by putting the following under
<dependencies> in your
<dependency> <groupId>edu.cmu.ml.rtw</groupdId> <artifactId>pra_2.11</artifactId> <version>3.3</version> </dependency>
Most recent PRA version: 3.3
See the changelog in the code's README for more info on what's new in each version.
Please note that in between versions the parameters available (and how they are specified) may change. If you are seeing configuration errors, or errors when processing parameters, you could either look in the code to see what the available options are (I've tried to make those relatively easy to find...), or you should be sure you have a tagged version of the code checked out. The documentation on this site reflects version 3.0. So, yes, this documentation is out of date, though some of it is still applicable. You can also look at the example directory in the github repo for example experiment configurations that should be working. If you have trouble, just ask me for help.
This code makes use of a number of other libraries that are distributed under various open source licenses (notably the Apache License and the Common Public License). You can see those dependencies listed in the pom.xml file. The code under the src/ directory is distributed under the terms of the GNU General Public License, version 3 (or, at your choosing, any later version of that license). You can find the text of that license here.