Learn discrete Bayesian network classifiers from data.

Description

State-of-the-art algorithms for learning discrete Bayesian network classifiers from data, with functions prediction, model evaluation and inspection.

Details

The learn more about the package, start with the vignettes: browseVignettes(package = "bnclassify"). The following is a list of available functionalities:

Structure learning algorithms:

• nb: Naive Bayes (Minsky, 1961)

• tan_cl: Chow-Liu's algorithm for one-dependence estimators (CL-ODE) (Friedman et al., 1997)

• fssj: Forward sequential selection and joining (FSSJ) (Pazzani, 1996)

• bsej: Backward sequential elimination and joining (BSEJ) (Pazzani, 1996)

• tan_hc: Hill-climbing tree augmented naive Bayes (TAN-HC) (Keogh and Pazzani, 2002)

• tan_hcsp: Hill-climbing super-parent tree augmented naive Bayes (TAN-HCSP) (Keogh and Pazzani, 2002)

• aode: Averaged one-dependence estimators (AODE) (Webb et al., 2005)

Parameter learning methods (lp):

• Bayesian and maximum likelihood estimation

• Weighting attributes to alleviate naive bayes' independence assumption (WANBIA) (Zaidi et al., 2013)

• Attribute-weighted naive Bayes (AWNB) (Hall, 2007)

• Model averaged naive Bayes (MANB) (Dash and Cooper, 2002)

Model evaluating:

• cv: Cross-validated estimate of accuracy

• logLik: Log-likelihood

• AIC: Akaike's information criterion (AIC)

• BIC: Bayesian information criterion (BIC)

Predicting:

• predict: Inference for complete and/or incomplete data (the latter through gRain)

Inspecting models:

• plot: Structure plotting (through igraph)

• print: Summary

• params: Access conditional probability tables

• nparams: Number of free parameters

• and more. See inspect_bnc_dag and inspect_bnc_bn.

Author(s)

Maintainer: Mihaljevic Bojan boki.mihaljevic@gmail.com [copyright holder]

Authors:

• Bielza Concha mcbielza@fi.upm.es

• Larranaga Pedro pedro.larranaga@fi.upm.es

Other contributors:

• Wickham Hadley (some code extracted from memoise package) [contributor]

References

Bielza C and Larranaga P (2014), Discrete Bayesian network classifiers: A survey. ACM Computing Surveys, 47(1), Article 5.

Dash D and Cooper GF (2002). Exact model averaging with naive Bayesian classifiers. 19th International Conference on Machine Learning (ICML-2002), 91-98.

Friedman N, Geiger D and Goldszmidt M (1997). Bayesian network classifiers. Machine Learning, 29, pp. 131–163.

Zaidi NA, Cerquides J, Carman MJ, and Webb GI (2013) Alleviating naive Bayes attribute independence assumption by attribute weighting. Journal of Machine Learning Research, 14 pp. 1947–1988.

GI. Webb, JR Boughton, and Z Wang (2005) Not so naive bayes: Aggregating one-dependence estimators. Machine Learning, 58(1) pp. 5–24.

Hall M (2007). A decision tree-based attribute weighting filter for naive Bayes. Knowledge-Based Systems, 20(2), pp. 120-126.

Koegh E and Pazzani M (2002).Learning the structure of augmented Bayesian classifiers. In International Journal on Artificial Intelligence Tools, 11(4), pp. 587-601.

Koller D, Friedman N (2009). Probabilistic Graphical Models: Principles and Techniques. MIT Press.

Pazzani M (1996). Constructive induction of Cartesian product attributes. In Proceedings of the Information, Statistics and Induction in Science Conference (ISIS-1996), pp. 66-77

See Also

Useful links:

• https://github.com/bmihaljevic/bnclassify

• Report bugs at https://github.com/bmihaljevic/bnclassify/issues

Compute predictive accuracy.

Description

Compute predictive accuracy.

Usage

accuracy(x, y)

Arguments

Examples

 
data(car)
nb <- bnc('nb', 'class', car, smooth = 1)
p <- predict(nb, car)
accuracy(p, car$class)

Learn an AODE ensemble.

Description

If there is a single predictor then returns a naive Bayes.

Usage

aode(class, dataset, features = NULL)

Arguments

Value

A bnc_aode or a bnc_dag (if returning a naive Bayes)

Checks if all columns in a data frame are factors.

Description

Checks if all columns in a data frame are factors.

Usage

are_factors(x)

Arguments

Returns TRUE is x is a valid probability distribution.

Description

Returns TRUE is x is a valid probability distribution.

Usage

are_pdists(x)

Convert to mlr.

Description

Convert a bnc_bn to a Learner object.

Usage

as_mlr(x, dag, id = "1")

Arguments

Examples

data(car)
nb <- bnc('nb', 'class', car, smooth = 1)
## Not run: library(mlr)
## Not run: nb_mlr <- as_mlr(nb, dag = FALSE, id = "ode_cl_aic")
## Not run: nb_mlr

Arcs that do not invalidate the k-DB structure

Description

Arcs that do not invalidate the k-DB structure

Usage

augment_kdb(kdbk)

Returns augmenting arcs that do not invalidate the k-DB.

Description

Returns augmenting arcs that do not invalidate the k-DB.

Usage

augment_kdb_arcs(bnc_dag, k)

Value

a character matrix. NULL if no arcs can be added.

Arcs that do not invalidate the tree-like structure

Description

Arcs that do not invalidate the tree-like structure

Usage

augment_ode(bnc_dag, ...)

Arguments

Returns augmenting arcs that do not invalidate the ODE.

Description

Returns augmenting arcs that do not invalidate the ODE.

Usage

augment_ode_arcs(bnc_dag)

Value

a character matrix. NULL if no arcs can be added.

Learn network structure and parameters.

Description

A convenience function to learn the structure and parameters in a single call. Must provide the name of the structure learning algorithm function; see bnclassify for the list.

Usage

bnc(
  dag_learner,
  class,
  dataset,
  smooth,
  dag_args = NULL,
  awnb_trees = NULL,
  awnb_bootstrap = NULL,
  manb_prior = NULL,
  wanbia = NULL
)

Arguments

Examples

data(car)
nb <- bnc('nb', 'class', car, smooth = 1)
nb_manb <- bnc('nb', 'class', car, smooth = 1, manb_prior = 0.3)
ode_cl_aic <- bnc('tan_cl', 'class', car, smooth = 1, dag_args = list(score = 'aic'))

Returns a c("bnc_aode", "bnc") object.

Description

Returns a c("bnc_aode", "bnc") object.

Usage

bnc_aode(models, class_var, features)

Fits an AODE model.

Description

Fits an AODE model.

Usage

bnc_aode_bns(x, fit_models)

Bayesian network classifier with structure and parameters.

Description

A Bayesian network classifier with structure and parameters. Returned by lp and bnc functions. You can use it to classify data (with predict). Can estimate its predictive accuracy with cv, plot its structure (with plot), print a summary to console (print), inspect it with functions documented in inspect_bnc_bn and inspect_bnc_dag, and convert it to mlr, grain, and graph objects –see as_mlr and grain_and_graph.

Examples

data(car)
tan <- bnc('tan_cl', 'class', car, smooth = 1)   
tan
p <- predict(tan, car)
head(p)
## Not run: plot(tan)
nparams(tan)

Bayesian network classifier structure.

Description

A Bayesian network classifier structure, returned by functions such as nb and tan_cl. You can plot its structure (with plot), print a summary to console (print), inspect it with functions documented in inspect_bnc_dag, and convert it to a graph object with grain_and_graph.

Examples

data(car)
nb <- tan_cl('class', car)   
nb
## Not run: plot(nb)
narcs(nb)

Return a bootstrap sub-sample.

Description

Return a bootstrap sub-sample.

Usage

bootstrap_ss(dataset, proportion)

Arguments

Car Evaluation Data Set.

Description

Data set from the UCI repository: https://archive.ics.uci.edu/ml/datasets/Car+Evaluation.

Format

A data.frame with 7 columns and 1728 rows.

Source

https://goo.gl/GTXrCz

Checks if mlr attached.

Description

mlr must be attached because otherwise 'getMlrOptions()' in 'makeLearner' will not be found.

Usage

check_mlr_attached()

Compute the (conditional) mutual information between two variables.

Description

Computes the (conditional) mutual information between two variables. If z is not NULL then returns the conditional mutual information, I(X;Y|Z). Otherwise, returns mutual information, I(X;Y).

Usage

cmi(x, y, dataset, z = NULL, unit = "log")

Arguments

Details

I(X;Y|Z) = H(X|Z) + H(Y|Z) - H(X,Y,Z) - H(Z), where H() is Shannon's entropy.

Examples

data(car)
cmi('maint', 'class', car)

Returns the conditional mutual information three variables.

Description

Returns the conditional mutual information three variables.

Usage

cmi_table(xyz_freqs, unit = "log")

Returns a complete unweighted graph with the given nodes.

Description

Returns a complete unweighted graph with the given nodes.

Usage

complete_graph(nodes)

Arguments

Value

a graphNEL object.

Computes the conditional log-likelihood of the model on the provided data.

Description

Computes the conditional log-likelihood of the model on the provided data.

Usage

compute_cll(x, dataset)

Computes log-likelihood of the model on the provided data.

Description

Computes log-likelihood of the model on the provided data.

Usage

compute_ll(x, dataset)

Arguments

Compute WANBIA weights. Computes feature weights by optimizing conditional log-likelihood. Weights are bounded to [0, 1]. Implementation based on the original paper and the code provided at https://sourceforge.net/projects/rawnaivebayes.

Description

Compute WANBIA weights.

Computes feature weights by optimizing conditional log-likelihood. Weights are bounded to [0, 1]. Implementation based on the original paper and the code provided at https://sourceforge.net/projects/rawnaivebayes.

Usage

compute_wanbia_weights(class, dataset, return_optim_object = FALSE)

Arguments

Value

a named numeric vector

Get just form first dimension in their own cpt, not checking for consistency in others.

Description

Get just form first dimension in their own cpt, not checking for consistency in others.

Usage

cpt_vars_values(cpts)

Estimate predictive accuracy with stratified cross validation.

Description

Estimate predictive accuracy of a classifier with stratified cross validation. It learns the models from the training subsamples by repeating the learning procedures used to obtain x. It can keep the network structure fixed and re-learn only the parameters, or re-learn both structure and parameters.

Usage

cv(x, dataset, k, dag = TRUE, mean = TRUE)

Arguments

Value

A numeric vector of same length as x, giving the predictive accuracy of each classifier. If mean = FALSE then a matrix with k rows and a column per each classifier in x.

Examples

data(car)
nb <- bnc('nb', 'class', car, smooth = 1) 
# CV a single classifier
cv(nb, car, k = 10) 
nb_manb <- bnc('nb', 'class', car, smooth = 1, manb_prior = 0.5) 
cv(list(nb=nb, manb=nb_manb), car, k = 10)
# Get accuracies on each fold
cv(list(nb=nb, manb=nb_manb), car, k = 10, mean = FALSE)
ode <- bnc('tan_cl', 'class', car, smooth = 1, dag_args = list(score = 'aic')) 
# keep structure fixed accross training subsamples
cv(ode, car, k = 10, dag = FALSE)

Get underlying graph. This should be exported.

Description

Get underlying graph. This should be exported.

Usage

dag(x)

Arguments

Direct an undirected graph.

Description

Starting from a root not, directs all arcs away from it and applies the same, recursively to its children and descendants. Produces a directed forest.

Usage

direct_forest(g, root = NULL)

Arguments

Value

A directed graph

Direct an undirected graph.

Description

The graph must be connected and the function produces a directed tree.

Usage

direct_tree(g, root = NULL)

Value

A graph. The directed tree.

Returns a contingency table over the variables.

Description

Each variable may be a character vector.

Usage

extract_ctgt(cols, dataset)

Details

Any rows with incomplete observations of the variables are ignored.

Compares all elements in a to b

Description

Compares all elements in a to b

Usage

fast_equal(a, b)

Arguments

Forget a memoized function.

Description

Forget a memoized function.

Usage

forget(f)

Author(s)

Hadley Wickham

Based on gRbase::ancestors()

Description

Based on gRbase::ancestors()

Usage

get_ancestors(node, families)

Return all but last element of x.

Description

If x is NULL returns NA not NULL

Usage

get_but_last(x)

Return last element of x.

Description

If x is NULL returns NA not NULL

Usage

get_last(x)

Assuming that the cpt is a leaf, returns 1 instead of a CPT entry when value missing

Description

Assuming that the cpt is a leaf, returns 1 instead of a CPT entry when value missing

Usage

get_log_leaf_entries(cpt, x)

Arguments

Get i-th element of x.

Description

If x is NULL returns NA not NULL

Usage

get_null_safe(x, i)

Convert to igraph and gRain.

Description

Convert a bnc_dag to igraph and grain objects.

Usage

as_igraph(x)

as_grain(x)

Arguments

Functions

• as_igraph(): Convert to a graphNEL.

• as_grain(): Convert to a grain.

Examples

data(car)
nb <- bnc('nb', 'class', car, smooth = 1)
# Requires the grain and igraph packages installed
## Not run: g <- as_grain(nb)
## Not run: gRain::querygrain.grain(g)$buying

Add edges Does not allow edges among adjacent nodes

Description

Add edges Does not allow edges among adjacent nodes

Usage

graph_add_edges(from, to, g)

connected_components

Description

connected_components

Usage

graph_connected_components(g)

Arguments

Finds adjacent nodes. Has not been tested much

Description

Finds adjacent nodes. Has not been tested much

Usage

graph_get_adjacent(node, g)

Checks whether nodes are adjacent

Description

Checks whether nodes are adjacent

Usage

graph_is_adjacent(from, to, g)

Returns an edge matrix with node names (instead of node indices).

Description

Returns an edge matrix with node names (instead of node indices).

Usage

graph_named_edge_matrix(x)

Value

A character matrix.

Subgraph. Only for a directed graph?

Description

Subgraph. Only for a directed graph?

Usage

graph_subgraph(nodes, g)

Arguments

Merges multiple disjoint graphs into a single one.

Description

Merges multiple disjoint graphs into a single one.

Usage

graph_union(g)

Arguments

Value

A graph

Learn Bayesian network classifiers in a a greedy wrapper fashion.

Description

Greedy wrapper algorithms for learning Bayesian network classifiers. All algorithms use cross-validated estimate of predictive accuracy to evaluate candidate structures.

Usage

fssj(class, dataset, k, epsilon = 0.01, smooth = 0, cache_reset = NULL)

bsej(class, dataset, k, epsilon = 0.01, smooth = 0, cache_reset = NULL)

tan_hc(class, dataset, k, epsilon = 0.01, smooth = 0, cache_reset = NULL)

kdb(
  class,
  dataset,
  k,
  kdbk = 2,
  epsilon = 0.01,
  smooth = 0,
  cache_reset = NULL
)

tan_hcsp(class, dataset, k, epsilon = 0.01, smooth = 0, cache_reset = NULL)

Arguments

Value

A bnc_dag object.

References

Pazzani M (1996). Constructive induction of Cartesian product attributes. In Proceedings of the Information, Statistics and Induction in Science Conference (ISIS-1996), pp. 66-77

Koegh E and Pazzani M (2002).Learning the structure of augmented Bayesian classifiers. In International Journal on Artificial Intelligence Tools, 11(4), pp. 587-601.

Examples

data(car)
tanhc <- tan_hc('class', car, k = 5, epsilon = 0)  
## Not run: plot(tanhc)
  

Identifies all depths at which the features of a classification tree are tested.

Description

Identifies all depths at which the features of a classification tree are tested.

Usage

identify_all_testing_depths(tree)

Arguments

Value

a numeric vector. The names are the names of the variables.

Identifies the lowest (closest to root) depths at which the features of a classification tree are tested.

Description

Identifies the lowest (closest to root) depths at which the features of a classification tree are tested.

Usage

identify_min_testing_depths(tree)

Inspect a Bayesian network classifier (with structure and parameters).

Description

Functions for inspecting a bnc_bn object. In addition, you can query this object with the functions documented in inspect_bnc_dag.

Usage

nparams(x)

manb_arc_posterior(x)

awnb_weights(x)

params(x)

values(x)

classes(x)

Arguments

Functions

• nparams(): Returns the number of free parameters in the model.

• manb_arc_posterior(): Returns the posterior of each arc from the class according to the MANB method.

• awnb_weights(): Returns the AWNB feature weights.

• params(): Returns the list of CPTs, in the same order as vars.

• values(): Returns the possible values of each variable, in the same order as vars.

• classes(): Returns the possible values of the class variable.

Examples

 
data(car)
nb <- bnc('nb', 'class', car, smooth = 1)
nparams(nb)
nb <- bnc('nb', 'class', car, smooth = 1, manb_prior = 0.5)
manb_arc_posterior(nb)
nb <- bnc('nb', 'class', car, smooth = 1, awnb_bootstrap = 0.5)
awnb_weights(nb)

Inspect a Bayesian network classifier structure.

Description

Functions for inspecting a bnc_dag object.

Usage

class_var(x)

features(x)

vars(x)

families(x)

modelstring(x)

feature_families(x)

narcs(x)

is_semi_naive(x)

is_anb(x)

is_nb(x)

is_ode(x)

Arguments

Functions

• class_var(): Returns the class variable.

• features(): Returns the features.

• vars(): Returns all variables (i.e., features + class).

• families(): Returns the family of each variable.

• modelstring(): Returns the model string of the network in bnlearn format (adding a space in between two families).

• feature_families(): Returns the family of each feature.

• narcs(): Returns the number of arcs.

• is_semi_naive(): Returns TRUE if x is a semi-naive Bayes.

• is_anb(): Returns TRUE if x is an augmented naive Bayes.

• is_nb(): Returns TRUE if x is a naive Bayes.

• is_ode(): Returns TRUE if x is a one-dependence estimator.

Examples

data(car)
nb <- bnc('nb', 'class', car, smooth = 1)
narcs(nb)
is_ode(nb)

Is it memoized?

Description

Is it memoized?

Usage

is.memoised(f)

Author(s)

Hadley Wickham

Is it en AODE?

Description

Is it en AODE?

Usage

is_aode(x)

Learn the parameters of a Bayesian network structure.

Description

Learn parameters with maximum likelihood or Bayesian estimation, the weighting attributes to alleviate naive bayes' independence assumption (WANBIA), attribute weighted naive Bayes (AWNB), or the model averaged naive Bayes (MANB) methods. Returns a bnc_bn.

Usage

lp(
  x,
  dataset,
  smooth,
  awnb_trees = NULL,
  awnb_bootstrap = NULL,
  manb_prior = NULL,
  wanbia = NULL
)

Arguments

Details

lp learns the parameters of each local distribution \theta_{ijk} = P(X_i = k \mid \mathbf{Pa}(X_i) = j) as

\theta_{ijk} = \frac{N_{ijk} + \alpha}{N_{ ij \cdot } + r_i \alpha},

where N_{ijk} is the number of instances in dataset in which X_i = k and \mathbf{Pa}(X_i) = j, N_{ ij \cdot} = \sum_{k=1}^{r_i} N_{ijk}, r_i is the cardinality of X_i, and all hyperparameters of the Dirichlet prior equal to \alpha. \alpha = 0 corresponds to maximum likelihood estimation. Returns a uniform distribution when N_{ i j \cdot } + r_i \alpha = 0. With partially observed data, the above amounts to available case analysis.

WANBIA learns a unique exponent 'weight' per feature. They are computed by optimizing conditional log-likelihood, and are bounded with all w_i \in [0, 1]. For WANBIA estimates, set wanbia to TRUE.

In order to get the AWNB parameter estimate, provide either the awnb_bootstrap and/or the awnb_trees argument. The estimate is:

\theta_{ijk}^{AWNB} = \frac{\theta_{ijk}^{w_i}}{\sum_{k=1}^{r_i} \theta_{ijk}^{w_i}},

while the weights w_i are computed as

w_i = \frac{1}{M}\sum_{t=1}^M \sqrt{\frac{1}{d_{ti}}},

where M is the number of bootstrap samples from dataset and d_{ti} the minimum testing depth of X_i in an unpruned classification tree learned from the t-th subsample (d_{ti} = 0 if X_i is omitted from t-th tree).

The MANB parameters correspond to Bayesian model averaging over the naive Bayes models obtained from all 2^n subsets over the n features. To get MANB parameters, provide the manb_prior argument.

Value

A bnc_bn object.

References

Hall M (2004). A decision tree-based attribute weighting filter for naive Bayes. Knowledge-based Systems, 20(2), 120-126.

Dash D and Cooper GF (2002). Exact model averaging with naive Bayesian classifiers. 19th International Conference on Machine Learning (ICML-2002), 91-98.

Pigott T D (2001) A review of methods for missing data. Educational research and evaluation, 7(4), 353-383.

Examples

data(car)
nb <- nb('class', car)
# Maximum likelihood estimation
mle <- lp(nb, car, smooth = 0)
# Bayesian estimaion
bayes <- lp(nb, car, smooth = 0.5)
# MANB
manb <- lp(nb, car, smooth = 0.5, manb_prior = 0.5)
# AWNB
awnb <- lp(nb, car, smooth = 0.5, awnb_trees = 10)

Learns a unpruned rpart recursive partition.

Description

Learns a unpruned rpart recursive partition.

Usage

learn_unprunned_tree(dataset, class)

Returns pairwise component of ODE (penalized) log-likelihood scores. In natural logarithms.

Description

Returns pairwise component of ODE (penalized) log-likelihood scores. In natural logarithms.

Usage

local_ode_score_contrib(x, y, class, dataset)

Normalize log probabilities.

Description

Uses the log-sum-exp trick.

Usage

log_normalize(lp)

References

Murphy KP (2012). Machine learning: a probabilistic perspective. The MIT Press. pp. 86-87.

Compute (penalized) log-likelihood.

Description

Compute (penalized) log-likelihood and conditional log-likelihood score of a bnc_bn object on a data set. Requires a data frame argument in addition to object.

Usage

## S3 method for class 'bnc_bn'
AIC(object, ...)

## S3 method for class 'bnc_bn'
BIC(object, ...)

## S3 method for class 'bnc_bn'
logLik(object, ...)

cLogLik(object, ...)

Arguments

Details

log-likelihood = log P(\mathcal{D} \mid \theta),

Akaike's information criterion (AIC) = log P(\mathcal{D} \mid \theta) - \frac{1}{2} |\theta|,

The Bayesian information criterion (BIC) score: = log P(\mathcal{D} \mid \theta) - \frac{\log N}{2} |\theta|,

where |\theta| is the number of free parameters in object, \mathcal{D} is the data set and N is the number of instances in \mathcal{D}.

cLogLik computes the conditional log-likelihood of the model.

Examples

data(car)
nb <- bnc('nb', 'class', car, smooth = 1)
logLik(nb, car)   
AIC(nb, car)
BIC(nb, car)
cLogLik(nb, car)   

makeRLearner. Auxiliary mlr function.

Description

makeRLearner. Auxiliary mlr function.

Usage

makeRLearner.bnc()

Returns a function to compute negative conditional log-likelihood given feature weights

Description

Returns a function to compute negative conditional log-likelihood given feature weights

Usage

make_cll(class_var, dataset)

Returns a function to compute the gradient of negative conditional log-likelihood with respect to feature weights

Description

Returns a function to compute the gradient of negative conditional log-likelihood with respect to feature weights

Usage

make_cll_gradient(class_var, dataset)

Assigns instances to the most likely class.

Description

Ties are resolved randomly.

Usage

map(pred)

Arguments

Value

a factor with the same levels as the class variable.

Returns the undirected augmenting forest.

Description

Uses Kruskal's algorithm to find the augmenting forest that maximizes the sum of pairwise weights. When the weights are class-conditional mutual information this forest maximizes the likelihood of the tree-augmented naive Bayes network.

Usage

max_weight_forest(g)

Arguments

Details

If g is not connected than this will return a forest; otherwise it is a tree.

Value

A graph. The maximum spanning forest.

References

Friedman N, Geiger D and Goldszmidt M (1997). Bayesian network classifiers. Machine Learning, 29, pp. 131–163.

Murphy KP (2012). Machine learning: a probabilistic perspective. The MIT Press. pp. 912-914.

Memoise a function.

Description

Based on Hadley Wickham's memoise package. Assumes that argument to f is a character vector.

Usage

memoise_char(f)

Arguments

Details

This function is a slightly modified version of memoise to avoid the use of digest. The rest functions copied as is from memoise.

Author(s)

Hadley Wickham, Bojan Mihaljevic

Learn a naive Bayes network structure.

Description

Learn a naive Bayes network structure.

Usage

nb(class, dataset = NULL, features = NULL)

Arguments

Value

A bnc_dag object.

Examples

 
data(car)
nb <- nb('class', car)   
nb2 <- nb('class', features = letters[1:10])
## Not run: plot(nb2)

Returns a naive Bayes structure

Description

Returns a naive Bayes structure

Usage

nb_dag(class, features)

Make a new cache.

Description

Make a new cache.

Usage

new_cache()

Author(s)

Hadley Wickham

Provide an acyclic ordering (i.e., a topological sort).

Description

Provide an acyclic ordering (i.e., a topological sort).

Usage

order_acyclic(families)

References

Beng-Jensen and Gutin, 2007, page 14.

Plot the structure.

Description

If node labels are to small to be viewed properly, you may fix label fontsize with argument fontsize. Also, you may try multiple different layouts.

Usage

## S3 method for class 'bnc_dag'
plot(x, y, layoutType = "dot", fontsize = NULL, ...)

Arguments

Examples

 

# Requires the igraph package to be installed.
data(car)
nb <- nb('class', car)
nb <- nb('class', car)
## Not run: plot(nb)
## Not run: plot(nb, fontsize = 20)
## Not run: plot(nb, layoutType = 'circo')
## Not run: plot(nb, layoutType = 'fdp')
## Not run: plot(nb, layoutType = 'osage')
## Not run: plot(nb, layoutType = 'twopi')
## Not run: plot(nb, layoutType = 'neato')

Predicts class labels or class posterior probability distributions.

Description

Predicts class labels or class posterior probability distributions.

Usage

## S3 method for class 'bnc_fit'
predict(object, newdata, prob = FALSE, ...)

Arguments

Details

Ties are resolved randomly. Inference is much slower if newdata contains NAs.

Value

If prob=FALSE, then returns a length-N factor with the same levels as the class variable in x, where N is the number of rows in newdata. Each element is the most likely class for the corresponding row in newdata. If prob=TRUE, returns a N by C numeric matrix, where C is the number of classes; each row corresponds to the class posterior of the instance.

Examples

data(car)
nb <- bnc('nb', 'class', car, smooth = 1)
p <- predict(nb, car)
head(p)
p <- predict(nb, car, prob = TRUE)
head(p)

predictLearner. Auxiliary mlr function.

Description

predictLearner. Auxiliary mlr function.

Usage

predictLearner.bnc(.learner, .model, .newdata, ...)

Arguments

Print basic information about a classifier.

Description

Print basic information about a classifier.

Usage

## S3 method for class 'bnc_base'
print(x, ...)

Whether to do checks or not. Set TRUE to speed up debugging or building.

Description

Whether to do checks or not. Set TRUE to speed up debugging or building.

Usage

skip_assert()

Skip while testing to isolate errors

Description

Skip while testing to isolate errors

Usage

skip_testing()

Returns a Superparent one-dependence estimator.

Description

Returns a Superparent one-dependence estimator.

Usage

spode(sp, features, class)

Arguments

Subset a 2D structure by a vector of column names.

Description

Not all colnames are necessarily in the columns of data; in that case this returns NA.

Usage

subset_by_colnames(colnames, data)

Arguments

Return nodes which can be superparents along with their possible children.

Description

Return nodes which can be superparents along with their possible children.

Usage

superparent_children(bnc_dag)

Value

list of search_state. NULL if no orphans

Learns a one-dependence estimator using Chow-Liu's algorithm.

Description

Learns a one-dependence Bayesian classifier using Chow-Liu's algorithm, by maximizing either log-likelihood, the AIC or BIC scores; maximizing log-likelihood corresponds to the well-known tree augmented naive Bayes (Friedman et al., 1997). When maximizing AIC or BIC the output might be a forest-augmented rather than a tree-augmented naive Bayes.

Usage

tan_cl(class, dataset, score = "loglik", root = NULL)

Arguments

Value

A bnc_dag object.

References

Friedman N, Geiger D and Goldszmidt M (1997). Bayesian network classifiers. Machine Learning, 29, pp. 131–163.

Examples

data(car)
ll <- tan_cl('class', car, score = 'loglik')   
## Not run: plot(ll)
ll <- tan_cl('class', car, score = 'loglik', root = 'maint')   
## Not run: plot(ll)
aic <- tan_cl('class', car, score = 'aic')   
bic <- tan_cl('class', car, score = 'bic')   

trainLearner. Auxiliary mlr function.

Description

trainLearner. Auxiliary mlr function.

Usage

trainLearner.bnc(.learner, .task, .subset, .weights, ...)

Arguments

Congress Voting Data Set.

Description

Data set from the UCI repository https://archive.ics.uci.edu/ml/datasets/Congressional+Voting+Records.

Format

A data.frame with 17 columns and 435 rows.

Source

https://goo.gl/GTXrCz