• Home Page
• Documentation
• Examples
• Research Notes
• Small Simulation Studies
• Bayesian Network Repository

• About the Author

Creating and manipulating objects

• Creating Bayesian network structures
  1. Creating an empty network
  2. Creating a saturated network
  3. Creating a network structure
     1. With a specific arc set
     2. With a specific adjacency matrix
     3. With a specific model formula
  4. Creating one or more random network structures
     1. With a specified node ordering
     2. Sampling from the space of connected directed acyclic graphs with uniform probability
     3. Sampling from the space of the directed acyclic graphs with uniform probability
• Creating custom fitted Bayesian networks
  1. Creating custom fitted Bayesian networks using expert knowledge
     1. Discrete networks
     2. Continuous networks
     3. Hybrid networks (mixed continuous and discrete nodes)
  2. Creating custom fitted Bayesian networks using both data and expert knowledge
• Manipulating the nodes of a network structure
  1. Adding and removing nodes
  2. Renaming nodes

Structure learning

• Score functions: computing & comparing
  1. Loading the reference data sets
  2. Computing a network score
  3. Testing score equivalence
• Conditional independence tests
  1. Parametric tests
  2. Semiparametric Monte Carlo tests
  3. Monte Carlo permutation tests
• Whitelists and blacklists in structure learning
  1. Constraint-based structure learning algorithms
  2. Score-based structure learning algorithms
  3. Local search learning algorithms
  4. Hybrid structure learning algorithms
• Utilities to construct blacklists
  1. Constructing a blacklist from a topological ordering
  2. Constructing a blacklist from a partial ordering
  3. Constructing a blacklist to ensure a subset of nodes are disconnected from each other
• Using custom scores in structure learning
  1. Replicating an existing score
  2. Instrumenting network scores to debug them
  3. Extending existing network scores
  4. Implementing new network scores
     1. Mixed-Effects models as local distributions
     2. Additive Bayesian networks
     3. The oracle score
• Using custom tests in structure learning
  1. Replicating an existing test
  2. Instrumenting tests to debug them
  3. Extending existing conditional independence tests
  4. Implementing new conditional independence tests
     1. Ordinal logistic regressions as local distributions
     2. The oracle test

Parameter learning

• Fitting the parameters of a Bayesian network
  1. Learning the network structure
  2. Setting the direction of undirected arcs
  3. Fitting the parameters (Maximum Likelihood estimates)
     1. Discrete data
     2. Continuous data
     3. Hybrid data (mixed discrete and continuous)
  4. Fitting the parameters (Bayesian Posterior estimates)
     1. Discrete data
  5. Fitting the parameters (Expectation-Maximization estimates)

Model validation

• Comparing Bayesian network structures
  1. Summaries
  2. Structural distances
  3. Visual comparisons
• Bootstrap-based inference
  1. The general case
  2. Measuring arc strength
• Bayesian networks and cross-validation
  1. Choosing a Bayesian network learning strategy
     1. k-fold cross-validation
     2. Custom folds in cross-validation
     3. Hold-out cross-validation
  2. Comparing different network structures
  3. Cross-validation and predictive error
  4. Cross-validation and predictive correlation
• Entropy and Kullback-Leibler divergence

Inference

• Computing the log-likelihood of data for a Bayesian network
• Predicting new observations from a Bayesian network
  1. Predicting from the parents
  2. Predicting with Monte Carlo posterior inference
  3. Predicting with exact inference
  4. Predictions in Bayesian network classifiers
• Approximate Monte Carlo inference
  1. Evaluating conditional probability queries
  2. Sampling from conditional probability distributions
  3. Likelihood weighting
• Bayesian network classifiers
  1. Structure learning
  2. Parameter learning
  3. Prediction

Plotting

• Plotting networks with the Rgraphviz package
  1. Options directly exposed from Rgraphviz
  2. Highlighting nodes and arcs
  3. Refining a plot using Rgraphviz functions
     1. Plotting networks left-to-right instead of top-to-bottom
• Plotting arc strengths with Rgraphviz and lattice
  1. Plotting the distribution of arc strengths
  2. Plotting the network structure together with the arc strengths
• Plotting networks and marginal distributions with the Rgraphviz package
  1. Discrete networks
  2. Continuous networks
  3. Hybrid networks

Handling of missing data

• Preprocessing data with missing values
  1. Discretizing data
  2. Removing highly-correlated variables
• Imputing missing values from a Bayesian network
  1. Imputing from the parents
  2. Imputing with Monte Carlo posterior inference
  3. Imputing with exact inference
• Local discovery algorithms for structure learning from data with missing values
• Constraint-based structure learning from data with missing values
• Score-based structure learning from data with missing values
  1. The Structural Expectation-Maximization (Structural EM) algorithm
     1. With partially observed variables
     2. With completely unobserved (latent) variables
  2. The Node-Average Likelihood
• Parameter learning from data with missing values
  1. Parameter estimators for complete data
  2. The Expectation-Maximization (EM) algorithm
• Bootstrap and cross-validation from data with missing values
• Log-likelihood of data with missing values
• Predicting from incomplete data

Interfacing with other R packages

• Interfacing with the deal R package
  1. Exporting a network structure to deal
  2. Importing a network structure from deal
• Interfacing with the pcalg R package
  1. Exporting a network structure to pcalg
  2. Importing a network structure from pcalg
• Interfacing with the parallel R package
  1. Parallel structure learning
  2. Parallel parameter learning
  3. Parallel cross-validation
  4. Parallel bootstrap
  5. Parallel approximate inference
• Interfacing with the graph R package
  1. Exporting a network structure to graph
  2. Importing a network structure from graph
• Interfacing with the igraph R package
  1. Exporting a network structure to igraph
  2. Importing a network structure from igraph
• Interfacing with the gRain R package
  1. Exporting a fitted Bayesian network to gRain
  2. Importing a fitted Bayesian network from gRain

Interfacing with other software packages

• Exporting networks to DOT files

Extended examples

• bnlearn: Practical Bayesian Networks in R (Tutorial at the useR! conference in Toulouse, 2019)
  1. A Quick introduction
     1. Bayesian networks
        1. Definitions
        2. Learning
        3. Inference
     2. The bnlearn package
  2. A Bayesian network analysis of malocclusion data
     1. The data
     2. Preprocessing and exploratory data analysis
  3. Model #1: a static Bayesian network as a difference model
     1. Learning the Bayesian network
        1. Learning the structure
        2. Learning the parameters
     2. Model validation
        1. Predictive accuracy
        2. Confirming with expert knowledge
  4. Model #2: a dynamic Bayesian network
     1. Learning the structure
     2. Model averaging in structure learning
     3. Learning the parameters
     4. Model validation and inference
• bnlearn: Learning Bayesian Networks from Incomplete Data (Advanced Data Science I course, Osaka University, 2023)
  1. A Quick introduction
     1. Bayesian networks
        1. Definitions
        2. Learning
        3. Inference
        4. Equivalence classes
        5. Causal interpretation of BNs
     2. Missing data
  2. Bayesian networks for missing data
     1. Parameter learning
        1. Using locally-complete data
        2. The Expectation-Maximisation algorithm
     2. Structure learning
        1. Using locally-complete data
        2. The Expectation-Maximisation algorithm
        3. The PC algorithm
     3. Inference
        1. Imputation
        2. Prediction