mmhc: The max-min-hill-climbing algorithm

In BauerMichael/max-min-hill-climbing-algorithm: The max-min-hill-climbing algorithm

Description

This algorithm reconstructs Bayesian Networks from observational data. Therefor it first builds the skeleton of the DAG (directed acyclic graph) with the max-min parents and children (mmpc) algorithm. Afterwards it directs the edges between the vertices with the Bayesian Dirichlet likelihood-equivalence uniform (BDeu) score. For more information on that read the report appended or "The max-min hill-climbing Bayesian network structure learning algorithm", by Ioannis Tsamardinos, Laura E. Brown & Constantin F. Aliferis.

Details

Package:	mmhc
Type:	Package
Version:	1.0
Date:	2014-09-06
License:	No License
R functions	----------------------
student(int x)	returns a data frame with x observations; one of the two self made examples
rainy(int x)	returns a data frame with x observations; one of the two self made examples
mmhc(data.frame x)	input: the observed data x (a data frame); calculates everything and makes a plot
C++ methods	----------------------
C <- new(MMHC, data.frame)	initalizes the mmhc class
C$mmpc()	executes the MMPC algorithm
C$mmhc()	executes the MMHC algorithm
C$pc()	this C++-methods returns the PC set
C$adjMat()	this C++-methods returns the adjacency matrix
C$score()	this C++-methods returns the score of the reconstructed graph
C$mat()	this C++-methods returns the data frame converted into an integer matrix

Author(s)

Maintainer: Michael Bauer <[e-mail]>

References

"The max-min hill-climbing Bayesian network structure learning algorithm", by Ioannis Tsamardinos, Laura E. Brown & Constantin F. Aliferis.
"A Scoring Function for Learning Bayesian Networks based on Mutual Information and Conditional Independence Tests", by Luis M. de Campos.

See Also

http://www.dsl-lab.org/supplements/mmhc_paper/paper_online.pdf
bnlearn-package

Examples

    
    ## Call the function
    
    data <- student(1000) # as above
    mmhc(data) # gives you the plot of the graph (no return value)
    
    
    ## Manuel Workflow: With this you are able to reconstruct the graph step by step
    
    library(Rcpp) # load Rcpp package
    library(igraph) # load igraph package
    data <- student(1000) # initalize the underlying example with 1000 observations
    C <- new(MMHC, data) # initalize the class object
    C$mmpc() # first reconstruct the skeleton (max-min parents and children algorithm)
    C$pc() # returns the PC set. It is a list where the n-th list element stands for the n-th node in your graph. The elements of one node are the parents/children of the node.
    C$mmhc() # set the edges (BDeu score)
    C$adjMat() # returns the adjacency matrix
    C$score() # returns the score of the graph
    plotObj <- graph.adjacency(C$adjMat()) # makes a plotable object with the igraph package
    plot(plotObj) # plots the object

BauerMichael/max-min-hill-climbing-algorithm documentation built on May 5, 2019, 10:33 a.m.