Graphical Models - Import

A Java library for performing inference in graphical models (Bayesian and Markov networks) and learning distribution parameters.

License

License

GroupId

GroupId

com.github.thorbenlindhauer
ArtifactId

ArtifactId

graphmod-import
Last Version

Last Version

0.1.0
Release Date

Release Date

Type

Type

jar
Description

Description

Graphical Models - Import
A Java library for performing inference in graphical models (Bayesian and Markov networks) and learning distribution parameters.

Download graphmod-import

How to add to project

<!-- https://jarcasting.com/artifacts/com.github.thorbenlindhauer/graphmod-import/ -->
<dependency>
    <groupId>com.github.thorbenlindhauer</groupId>
    <artifactId>graphmod-import</artifactId>
    <version>0.1.0</version>
</dependency>
// https://jarcasting.com/artifacts/com.github.thorbenlindhauer/graphmod-import/
implementation 'com.github.thorbenlindhauer:graphmod-import:0.1.0'
// https://jarcasting.com/artifacts/com.github.thorbenlindhauer/graphmod-import/
implementation ("com.github.thorbenlindhauer:graphmod-import:0.1.0")
'com.github.thorbenlindhauer:graphmod-import:jar:0.1.0'
<dependency org="com.github.thorbenlindhauer" name="graphmod-import" rev="0.1.0">
  <artifact name="graphmod-import" type="jar" />
</dependency>
@Grapes(
@Grab(group='com.github.thorbenlindhauer', module='graphmod-import', version='0.1.0')
)
libraryDependencies += "com.github.thorbenlindhauer" % "graphmod-import" % "0.1.0"
[com.github.thorbenlindhauer/graphmod-import "0.1.0"]

Dependencies

compile (2)

Group / Artifact Type Version
com.github.thorbenlindhauer : graphmod-inference-engine jar 0.1.0
org.codehaus.woodstox : woodstox-core-asl jar 4.4.1

test (2)

Group / Artifact Type Version
junit : junit jar 4.11
org.assertj : assertj-core jar 1.6.1

Project Modules

There are no modules declared in this project.

graphical-models

graphical-models is a Java library of exact and approximate inference algorithms in Bayesian and Markov networks.

Inference can be performed in networks with factors of types:

  • Discrete factors
  • Canonical Gaussian factors (including Gaussiand and conditional linear Gaussian distributions)

It provides implementations of the following algorithms:

  • Variable Elimination
  • Clique Tree Inference
  • Loopy Belief Propagation
  • Expectation Propagation

Networks are specified in terms of cluster graphs. A cluster, a node in the cluster graph, may consist of any number of factors.

Setup

Maven dependency:

<dependency>
  <groupId>com.github.thorbenlindhauer</groupId>
  <artifactId>graphmod-inference-engine</artifactId>
  <version>0.1.0</version>
</dependency>

How to Use

Yet to come.

Variable Elimination

The strategy for determining the elimination order can be exchanged. The following default implementations exist:

  • Min-fill strategy: Eliminate variables in the order that introduces the least number of fill edges to the factor graph

Clique Tree Inference

Implementations of the sum product and the belief update schema exist. Clique trees can be generated for a given factor graph based on a variable elimination strategy.

Loopy Belief Propagation

Strategies for defining message order and determining calibration in the graph can be exchanged.

Expectation Propagation

Expectation propagation is complementary to clique tree inference and loopy belief propagation. It serves as a mean for approximation when factor products cannot be performed in closed form. The strategy of approximating a set of factors that form a cluster is configurable and extensible. The following implementations exist:

  • Univariate truncated Gaussians can be approximated in Gaussian networks

Defining Models

A fluent builder for defining factors, grouping factors to clusters, and creating a graph out of them exist. A rudimentary import for graphs specified in the XMLBIF format is provided in addition.

Reading

This library based on the concepts described in the book Probabilistic Graphical Models by Daphne Koller and Nir Friedmann:

Koller, Daphne, and Nir Friedman Probabilistic graphical models: principles and techniques MIT press, 2009

Versions

Version
0.1.0