Index  Topics 
bnlearnpackage {bnlearn}  R Documentation 
Bayesian network structure learning, parameter learning and inference
Description
Bayesian network structure learning (via constraintbased, scorebased and hybrid algorithms), parameter learning (via ML and Bayesian estimators) and inference.
Details
Package:  bnlearn 
Type:  Package 
Version:  4.520190307 
Date:  20190307 
License:  GPLv2 or later 
This package implements some algorithms for learning the structure of Bayesian networks.
Constraintbased algorithms, also known as conditional independence learners, are all optimized derivatives of the Inductive Causation algorithm (Verma and Pearl, 1991). These algorithms use conditional independence tests to detect the Markov blankets of the variables, which in turn are used to compute the structure of the Bayesian network.
Scorebased learning algorithms are general purpose heuristic optimization algorithms which rank network structures with respect to a goodnessoffit score.
Hybrid algorithms combine aspects of both constraintbased and scorebased algorithms, as they use conditional independence tests (usually to reduce the search space) and network scores (to find the optimal network in the reduced space) at the same time.
Several functions for parameter estimation, parametric inference, bootstrap, crossvalidation and stochastic simulation are available. Furthermore, advanced plotting capabilities are implemented on top of the Rgraphviz and lattice packages.
Available ConstraintBased Learning Algorithms

PC (
pc.stable
), a modern implementation of the first practical constraintbased structure learning algorithm. 
GrowShrink (
gs
): based on the GrowShrink Markov Blanket, the first (and simplest) Markov blanket detection algorithm used in a structure learning algorithm. 
Incremental Association (
iamb
): based on the Markov blanket detection algorithm of the same name, which is based on a twophase selection scheme (a forward selection followed by an attempt to remove false positives). 
Fast Incremental Association (
fast.iamb
): a variant of IAMB which uses speculative stepwise forward selection to reduce the number of conditional independence tests. 
Interleaved Incremental Association (
inter.iamb
): another variant of IAMB which uses forward stepwise selection to avoid false positives in the Markov blanket detection phase.
This package includes two implementations of each algorithm: a vanilla implementation and a parallel that
requires a running cluster set up with the makeCluster
function from the parallel package.
The computational complexity of these algorithms is polynomial in the number of tests, usually O(N^2) (but superexponential in the worst case scenario), where N is the number of variables. Execution time scales linearly with the size of the data set.
Available Scorebased Learning Algorithms

HillClimbing (
hc
): a hill climbing greedy search on the space of the directed graphs. The optimized implementation uses score caching, score decomposability and score equivalence to reduce the number of duplicated tests. 
Tabu Search (
tabu
): a modified hillclimbing able to escape local optima by selecting a network that minimally decreases the score function.
Random restart with a configurable number of perturbing operations is implemented for both algorithms.
Available Hybrid Learning Algorithms

MaxMin HillClimbing (
mmhc
): a hybrid algorithm which combines the MaxMin Parents and Children algorithm (to restrict the search space) and the HillClimbing algorithm (to find the optimal network structure in the restricted space). 
Restricted Maximization (
rsmax2
): a more general implementation of the MaxMin HillClimbing, which can use any combination of constraintbased and scorebased algorithms.
Other (ConstraintBased) Local Discovery Algorithms
These algorithms learn the structure of the undirected graph underlying the Bayesian network, which is known as the skeleton of the network or the (partial) correlation graph. Therefore all the arcs are undirected, and no attempt is made to detect their orientation. They are often used in hybrid learning algorithms.

MaxMin Parents and Children (
mmpc
): a forward selection technique for neighbourhood detection based on the maximization of the minimum association measure observed with any subset of the nodes selected in the previous iterations. 
Hiton Parents and Children (
si.hiton.pc
): a fast forward selection technique for neighbourhood detection designed to exclude nodes early based on the marginal association. The implementation follows the SemiInterleaved variant of the algorithm. 
ChowLiu (
chow.liu
): an application of the minimumweight spanning tree and the information inequality. It learns the tree structure closest to the true one in the probability space. 
ARACNE (
aracne
): an improved version of the ChowLiu algorithm that is able to learn polytrees.
All these algorithms have two implementations (vanilla and parallel) like other constraintbased algorithms.
Bayesian Network Classifiers
The algorithms are aimed at classification, and favour predictive power over the ability to recover the correct network structure. The implementation in bnlearn assumes that all variables, including the classifiers, are discrete.

Naive Bayes (
naive.bayes
): a very simple algorithm assuming that all classifiers are independent and using the posterior probability of the target variable for classification. 
TreeAugmented Naive Bayes (
tree.bayes
): an improvement over naive Bayes, this algorithms uses ChowLiu to approximate the dependence structure of the classifiers.
Available (Conditional) Independence Tests
The conditional independence tests used in constraintbased algorithms in practice are statistical tests on the data set. Available tests (and the respective labels) are:

discrete case (categorical variables)

mutual information: an informationtheoretic distance measure. It's proportional to the loglikelihood ratio (they differ by a 2n factor) and is related to the deviance of the tested models. The asymptotic chisquare test (
mi
andmiadf
, with adjusted degrees of freedom), the Monte Carlo permutation test (mcmi
), the sequential Monte Carlo permutation test (smcmi
), and the semiparametric test (spmi
) are implemented. 
shrinkage estimator for the mutual information (
mish
): an improved asymptotic chisquare test based on the JamesStein estimator for the mutual information. 
Pearson's X^2: the classical Pearson's X^2 test for contingency tables. The asymptotic chisquare test (
x2
andx2adf
, with adjusted degrees of freedom), the Monte Carlo permutation test (mcx2
), the sequential Monte Carlo permutation test (smcx2
) and semiparametric test (spx2
) are implemented.


discrete case (ordered factors)

JonckheereTerpstra: a trend test for ordinal variables. The asymptotic normal test (
jt
), the Monte Carlo permutation test (mcjt
) and the sequential Monte Carlo permutation test (smcjt
) are implemented.


continuous case (normal variables)

linear correlation: Pearson's linear correlation. The exact Student's t test (
cor
), the Monte Carlo permutation test (mccor
) and the sequential Monte Carlo permutation test (smccor
) are implemented. 
Fisher's Z: a transformation of the linear correlation with asymptotic normal distribution. Used by commercial software (such as TETRAD II) for the PC algorithm (an R implementation is present in the
pcalg
package on CRAN). The asymptotic normal test (zf
), the Monte Carlo permutation test (mczf
) and the sequential Monte Carlo permutation test (smczf
) are implemented. 
mutual information: an informationtheoretic distance measure. Again it is proportional to the loglikelihood ratio (they differ by a 2n factor). The asymptotic chisquare test (
mig
), the Monte Carlo permutation test (mcmig
) and the sequential Monte Carlo permutation test (smcmig
) are implemented. 
shrinkage estimator for the mutual information (
migsh
): an improved asymptotic chisquare test based on the JamesStein estimator for the mutual information.


hybrid case (mixed discrete and normal variables)

mutual information: an informationtheoretic distance measure. Again it is proportional to the loglikelihood ratio (they differ by a 2n factor). Only the asymptotic chisquare test (
micg
) is implemented.

Available Network Scores
Available scores (and the respective labels) are:

discrete case (categorical variables)

the multinomial loglikelihood (
loglik
) score, which is equivalent to the entropy measure used in Weka. 
the Akaike Information Criterion score (
aic
). 
the Bayesian Information Criterion score (
bic
), which is equivalent to the Minimum Description Length (MDL) and is also known as Schwarz Information Criterion. 
the logarithm of the Bayesian Dirichlet equivalent score (
bde
), a score equivalent Dirichlet posterior density. 
the logarithm of the Bayesian Dirichlet sparse score (
bds
), a sparsityinducing Dirichlet posterior density (not score equivalent). 
the logarithm of the Bayesian Dirichlet score with Jeffrey's prior (not score equivalent).

the logarithm of the modified Bayesian Dirichlet equivalent score (
mbde
) for mixtures of experimental and observational data (not score equivalent). 
the logarithm of the locally averaged Bayesian Dirichlet score (
bdla
, not score equivalent). 
the logarithm of the K2 score (
k2
), a Dirichlet posterior density (not score equivalent).


continuous case (normal variables)

the multivariate Gaussian loglikelihood (
loglikg
) score. 
the corresponding Akaike Information Criterion score (
aicg
). 
the corresponding Bayesian Information Criterion score (
bicg
). 
a score equivalent Gaussian posterior density (
bge
).


hybrid case (mixed discrete and normal variables)

the conditional linear Gaussian loglikelihood (
loglikcg
) score. 
the corresponding Akaike Information Criterion score (
aiccg
). 
the corresponding Bayesian Information Criterion score (
biccg
).

Whitelist and Blacklist Support
All learning algorithms support arc whitelisting and blacklisting:

blacklisted arcs are never present in the graph.

arcs whitelisted in one direction only (i.e. A > B is whitelisted but B > A is not) have the respective reverse arcs blacklisted, and are always present in the graph.

arcs whitelisted in both directions (i.e. both A > B and B > A are whitelisted) are present in the graph, but their direction is set by the learning algorithm.
Any arc whitelisted and blacklisted at the same time is assumed to be whitelisted, and is thus removed from the blacklist.
In algorithms that learn undirected graphs, such as ARACNE and ChowLiu, arcs are treated as being blacklisted in both directions even if only one direction is listed in the blackist.
Author(s)
Marco Scutari
Istituto Dalle Molle di Studi sull'Intelligenza Artificiale (IDSIA)
References
Nagarajan R, Scutari M, Lebre S (2013). "Bayesian Networks in R with Applications in Systems Biology". Springer.
Scutari M (2010). "Learning Bayesian Networks with the bnlearn R Package". Journal of Statistical Software, 35(3):1–22.
Scutari M (20107). "Bayesian Network ConstraintBased Structure Learning Algorithms: Parallel and Optimized Implementations in the bnlearn R Package". Journal of Statistical Software, 77(2):1–20.
Koller D, Friedman N (2009). Probabilistic Graphical Models: Principles and Techniques. MIT Press.
Korb K, Nicholson AE (2010). Bayesian Artificial Intelligence. Chapman & Hall/CRC, 2nd edition.
Pearl J (1988). Probabilistic Reasoning in Intelligent Systems: Networks of Plausible Inference. Morgan Kaufmann.
Examples
library(bnlearn) data(learning.test) ## Simple learning # first try the GrowShrink algorithm res = gs(learning.test) # plot the network structure. plot(res) # now try the Incremental Association algorithm. res2 = iamb(learning.test) # plot the new network structure. plot(res2) # the network structures seem to be identical, don't they? all.equal(res, res2) # how many tests each of the two algorithms used? ntests(res) ntests(res2) ## Greedy search res = hc(learning.test) plot(res) ## Another simple example (Gaussian data) data(gaussian.test) # first try the GrowShrink algorithm res = gs(gaussian.test) plot(res) ## Blacklist and whitelist use # the arc B  F should not be there? blacklist = data.frame(from = c("B", "F"), to = c("F", "B")) blacklist res3 = gs(learning.test, blacklist = blacklist) plot(res3) # force E  F direction (E > F). whitelist = data.frame(from = c("E"), to = c("F")) whitelist res4 = gs(learning.test, whitelist = whitelist) plot(res4) # use both blacklist and whitelist. res5 = gs(learning.test, whitelist = whitelist, blacklist = blacklist) plot(res5) ## Debugging # use the debugging mode to see the learning algorithms # in action. res = gs(learning.test, debug = TRUE) res = hc(learning.test, debug = TRUE) # log the learning process for future reference. ## Not run: sink(file = "learninglog.txt") res = gs(learning.test, debug = TRUE) sink() ## End(Not run)
Index  Topics 