Learn R Programming

bnstruct (version 1.0)

learn.network: learn a network (structure and parameters) of a BN from a BNDataset.

Description

Learn a network (structure and parameters) of a BN from a BNDataset (see the Details section).

Usage

learn.network(x, ...)
"learn.network"(x, y = NULL, algo = "mmhc", scoring.func = "BDeu", initial.network = NULL, alpha = 0.05, ess = 1, bootstrap = FALSE, layering = c(), max.fanin.layers = NULL, max.fanin = num.variables(dataset), layer.struct = NULL, cont.nodes = c(), use.imputed.data = FALSE, use.cpc = TRUE, ...)
"learn.network"(x, algo = "mmhc", scoring.func = "BDeu", initial.network = NULL, alpha = 0.05, ess = 1, bootstrap = FALSE, layering = c(), max.fanin.layers = NULL, max.fanin = num.variables(dataset), layer.struct = NULL, cont.nodes = c(), use.imputed.data = FALSE, use.cpc = TRUE, ...)

Arguments

x
can be a BN or a BNDataset. If x is a BN, then also the dataset parameter must be given.
...
potential further arguments for methods.
y
a BNDataset object, to be provided only if x is a BN.
algo
the algorithm to use. Currently, one among: sm (Silander-Myllymaki), mmhc (Max-Min Hill Climbing, default) and sem (Structural Expectation Maximization).
scoring.func
the scoring function to use. Currently, one among BDeu, AIC, BIC.
initial.network
network srtructure to be used as starting point for structure search. Can take different values: a BN object, a matrix containing the adjacency matrix of the structure of the network, or the string random.chain to sample a random chain as starting point.
alpha
confidence threshold (only for mmhc).
ess
Equivalent Sample Size value.
bootstrap
TRUE to use bootstrap samples.
layering
vector containing the layers each node belongs to.
max.fanin.layers
matrix of available parents in each layer (only for sm).
max.fanin
maximum number of parents for each node (only for sm).
layer.struct
0/1 matrix for indicating which layers can contain parent nodes for nodes in a layer (only for mmhc).
cont.nodes
vector containing the index of continuous variables.
use.imputed.data
TRUE to learn the structure from the imputed dataset (if available, a check is performed). Default is to use raw dataset
use.cpc
(when using mmhc) compute Candidate Parent-and-Children sets instead of starting the Hill Climbing from an empty graph.

Value

new BN object with structure (DAG) and conditional probabilities as learnt from the given dataset.

Details

Learn the structure (the directed acyclic graph) of a BN object according to a BNDataset. We provide three algorithms in order to learn the structure of the network, that can be chosen with the algo parameter. The first is the Silander-Myllym\"aki (sm) exact search-and-score algorithm, that performs a complete evaluation of the search space in order to discover the best network; this algorithm may take a very long time, and can be inapplicable when discovering networks with more than 25--30 nodes. Even for small networks, users are strongly encouraged to provide meaningful parameters such as the layering of the nodes, or the maximum number of parents -- refer to the documentation in package manual for more details on the method parameters.

The second algorithm (and the default one) is the Max-Min Hill-Climbing heuristic (mmhc), that performs a statistical sieving of the search space followed by a greedy evaluation. It is considerably faster than the complete method, at the cost of a (likely) lower quality. Also note that in the case of a very dense network and lots of obsevations, the statistical evaluation of the search space may take a long time. Also for this algorithm there are parameters that may need to be tuned, mainly the confidence threshold of the statistical pruning.

The third method is the Structural Expectation-Maximization (sem) algorithm, for learning a network from a dataset with missing values. It iterates a sequence of Expectation-Maximization (in order to ``fill in'' the holes in the dataset) and structure learning from the guessed dataset, until convergence. The structure learning used inside SEM, due to computational reasons, is MMHC. Convergence of SEM can be controlled with the parameters struct.threshold and param.threshold, for the structure and the parameter convergence, respectively.

Search-and-score methods also need a scoring function to compute an estimated measure of each configuration of nodes. We provide three of the most popular scoring functions, BDeu (Bayesian-Dirichlet equivalent uniform, default), AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion). The scoring function can be chosen using the scoring.func parameter.

Then, the parameters of the network are learnt using MAP (Maximum A Posteriori) estimation (if not using bootstrap).

See documentation for learn.structure and learn.params for more informations.

See Also

learn.structure learn.params

Examples

Run this code
## Not run: 
# mydataset <- BNDataset("data.file", "header.file")
# 
# # starting from a BN
# net <- BN(mydataset)
# net <- learn.network(net, mydataset)
# 
# # start directly from the dataset
# net <- learn.network(mydataset)
# ## End(Not run)

Run the code above in your browser using DataLab