Replacement function for parameters inside DBNs

Description

Generic parameter replacement method for "dbn.fit" S3 objects. Calls bnlearn underneath.

Usage

## S3 replacement method for class 'dbn.fit'
x$name <- value

Arguments

Value

the modified network

Calculate the AIC of a dynamic Bayesian network

Description

Generic method for calculating the Akaike information criterion (AIC) of a "dbn" S3 object given some data. Calls bnlearn's AIC underneath.

Usage

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

Arguments

Value

the AIC score of the network

Calculate the AIC of a dynamic Bayesian network

Description

Generic method for calculating the Akaike information criterion (AIC) of a "dbn.fit" S3 object given some data. Calls bnlearn's AIC underneath.

Usage

## S3 method for class 'dbn.fit'
AIC(object, ..., k)

Arguments

Value

the AIC score of the network

Calculate the BIC of a dynamic Bayesian network

Description

Generic method for calculating the Bayesian information criterion (BIC) of a "dbn" S3 object given some data. Calls bnlearn's BIC underneath.

Usage

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

Arguments

Value

the BIC score of the network

Calculate the BIC of a dynamic Bayesian network

Description

Generic method for calculating the Bayesian information criterion (BIC) of a "dbn.fit" S3 object given some data. Calls bnlearn's BIC underneath.

Usage

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

Arguments

Value

the BIC score of the network

R6 class that defines causal lists in the PSO

Description

Constructor of the 'Causlist' class

Arguments

Details

The causal lists will be the base of the positions and the velocities in the pso part of the algorithm.

Value

A new 'causlist' object

Fields

cl

List of causal units

size

Size of the DBN

ordering

String vector defining the order of the nodes in a timeslice

R6 class that defines a Particle in the PSO algorithm

Description

Constructor of the 'Particle' class

Evaluate the score of the particle's position

Evaluate the score of the particle's position. Updates the local best if the new one is better.

Update the position of the particle with the velocity

Update the position of the particle given the constants after calculating the new velocity

Arguments

Details

A particle has a Position, a Velocity and a local best

Value

A new 'Particle' object

The score of the current position

Fields

ps

position of the particle

cl

velocity of the particle

lb

local best score obtained

lb_ps

local best position found

score

bnlearn score function used

R6 class that defines DBNs as causality lists

Description

Constructor of the 'causlist' class

Translate the causality list into a DBN network

Uses this object private causality list and transforms it into a DBN.

Add a velocity to the position

Given a Velocity object, add it to the current position.

Given another position, returns the velocity that gets this position to the other.

Return the static node ordering

This function takes as input a dbn and return the node ordering of the variables inside a timeslice. This ordering is needed to understand a causal list.

Translate a DBN into a causality list

This function takes as input a network from a DBN and transforms the structure into a causality list if it is a valid DBN. Valid DBNs have only inter-timeslice edges and only allow variables in t_0 to have parents.

Generates a random DBN valid for causality list translation

This function takes as input a list with the names of the nodes and the desired size of the network and returns a random DBN structure.

Fixes a DBN structure to make it suitable for causality list translation

This function takes as input a DBN structure and removes the intra-timeslice arcs and the arcs that end in a node not in t_0.

Arguments

Details

A causality list has a list with causal units, a size representing the Markovian order of the network and a specific node ordering.

Value

A new 'causlist' object

a dbn object

the ordering of the nodes in t_0

a causlist object

a random dbn structure

the fixed network

Fields

n_arcs

Number of arcs in the network

nodes

Names of the nodes in the network

R6 class that defines the PSO controller

Description

Constructor of the 'PsoCtrl' class

Getter of the cluster attribute

Transforms the best position found into a bn structure and returns it

Main function of the pso algorithm.

Initialize the particles for the algorithm to random positions and velocities.

Evaluate the particles and update the global best

Modify the PSO parameters after each iteration

Arguments

Details

The controller will encapsulate the particles and run the algorithm

Value

A new 'PsoCtrl' object

the cluster attribute

the size attribute

Fields

parts

list with all the particles in the algorithm

cl

cluster for the parallel computations

n_it

maximum number of iterations of the pso algorithm

in_cte

parameter that varies the effect of the inertia

gb_cte

parameter that varies the effect of the global best

lb_cte

parameter that varies the effect of the local best

b_ps

global best position found

b_scr

global best score obtained

r_probs

vector that defines the range of random variation of gb_cte and lb_cte

cte

boolean that defines whether the parameters remain constant or vary as the execution progresses

in_var

decrement of the inertia each iteration

gb_var

increment of the global best parameter each iteration

lb_var

increment of the local best parameter each iteration

R6 class that defines velocities affecting causality lists in the PSO

Description

Getter of the abs_op attribute.

return the number of operations that the velocity performs

Setter of the abs_op attribute. Intended for inside use only. This should be a 'protected' function in Java-like OOP, but there's no such thing in R6. This function should not be used from outside the package.

Randomizes the Velocity's directions. If the seed provided is NULL, no seed will be used.

Given a position, returns the velocity that gets this position to the other.

Add both velocities directions

Multiply the Velocity by a constant real number

This function multiplies the Velocity by a constant real number. It is non deterministic by definition. When calculating k*|V|, the result will be floored and bounded to the set [-max_op, max_op], where max_op is the maximum number of arcs that can be present in the network.

Arguments

Details

The velocities will be defined as a causality list where each element in a causal unit is a pair (v, node) with v being either 0, 1 or -1. 0 means that arc remained the same, 1 means that arc was added and -1 means that arc was deleted.

Fields

abs_op

Total number of operations 1 or -1 in the velocity

Replacement function for parameters inside DBNs

Description

Generic parameter replacement method for "dbn.fit" S3 objects. Calls bnlearn underneath.

Usage

## S3 replacement method for class 'dbn.fit'
x[[name]] <- value

Arguments

Value

the modified network

Returns a vector with the number of consecutive nodes in each level

Description

This method processes the vector of node levels to get the position of each node inside the level. E.g. c(1,1,1,2,2,3,4,4,5,5) turns into c(1,2,3,1,2,1,1,2,1,2)

Usage

acc_successions(nodes, res = NULL, prev = 0, acc = 0)

Arguments

Value

the vector of sub successions in each level

Adds the mu vector and sigma matrix as attributes to the bn.fit or dbn.fit object

Description

Adds the mu vector and sigma matrix as attributes to the bn.fit or dbn.fit object to allow performing exact MVN inference on both cases. It also adds the number of time slices of the net for future inference.

Usage

add_attr_to_fit(fit, size)

Arguments

Value

the fitted net with attributes

Check if two network structures are equal to each other

Description

Generic method for checking the equality of two "dbn" S3 objects. Calls bnlearn's all.equal underneath.

Usage

## S3 method for class 'equal.dbn'
all(target, current, ...)

Arguments

Value

boolean result of the comparison

Check if two fitted networks are equal to each other

Description

Generic method for checking the equality of two "dbn.fit" S3 objects. Calls bnlearn's all.equal underneath.

Usage

## S3 method for class 'equal.dbn.fit'
all(target, current, ...)

Arguments

Value

boolean result of the comparison

Performs approximate inference in a time slice of the dbn

Description

Given a bn.fit object and some variables, performs particle inference over such variables in the net for a given time slice.

Usage

approx_prediction_step(fit, variables, particles, n = 50)

Arguments

Value

the inferred particles

Performs approximate inference forecasting with the GDBN over a dataset

Description

Given a bn.fit object, the size of the net and a dataset, performs approximate forecasting with bnlearns cpdist function over the initial evidence taken from the dataset.

Usage

approximate_inference(dt, fit, obj_vars, ini, rep, len, num_p)

Arguments

Value

a list with the mu results of the forecast

Convert a network structure into a model string

Description

Generic method for converting a "dbn" S3 object into a string. Calls bnlearn's as.character underneath.

Usage

## S3 method for class 'dbn'
as.character(x, ...)

Arguments

Value

string representing the DBN model

Converts a single row data.table into a named vector

Description

Given a single row data.table, convert it into a named vector. Used in the mvn_inference, to obtain an evidence named vector. For this case, the data.table should contain only the evidence that we want to provide. If a named vector is provided instead of a data.table, nothing will be done and the named vector will be returned.

Usage

as_named_vector(dt)

Arguments

Value

a named vector with the values and the variable names

Experimental function that translates a natPosition vector into a DBN network.

Description

Experimental function that translates a natPosition vector into a DBN network.

Usage

bn_translate_exp(ps, ordering_raw, n_arcs, nodes)

Arguments

Value

a bn object

Calculate the mu vector from a fitted BN or DBN

Description

Given a "bn.fit" or a "dbn.fit" object, calculate the mu vector of the equivalent multivariate Gaussian distribution. Front end of a C++ function.

Usage

calc_mu(fit)

Arguments

Value

a named numeric vector of the means of each variable

Examples

dt_train <- dbnR::motor[200:2500]
net <- bnlearn::mmhc(dt_train)
fit <- bnlearn::bn.fit(net, dt_train, method = "mle-g")
mu <- dbnR::calc_mu(fit)

f_dt_train <- dbnR::fold_dt(dt_train, size = 2)
net <- dbnR::learn_dbn_struc(dt_train, size = 2)
fit <- dbnR::fit_dbn_params(net, f_dt_train)
mu <- dbnR::calc_mu(fit)

Calculate the mu vector of means of a Gaussian linear network. This is the C++ backend of the function.

Description

Calculate the mu vector of means of a Gaussian linear network. This is the C++ backend of the function.

Usage

calc_mu_cpp(fit, order)

Arguments

Value

the map with the nodes and their mu. Returns as a named numeric vector

Calculate the sigma covariance matrix from a fitted BN or DBN

Description

Given a "bn.fit" or a "dbn.fit" object, calculate the sigma covariance matrix of the equivalent multivariate Gaussian distribution. Front end of a C++ function.

Usage

calc_sigma(fit)

Arguments

Value

a named numeric covariance matrix of the nodes

Examples

dt_train <- dbnR::motor[200:2500]
net <- bnlearn::mmhc(dt_train)
fit <- bnlearn::bn.fit(net, dt_train, method = "mle-g")
sigma <- dbnR::calc_sigma(fit)

f_dt_train <- dbnR::fold_dt(dt_train, size = 2)
net <- dbnR::learn_dbn_struc(dt_train, size = 2)
fit <- dbnR::fit_dbn_params(net, f_dt_train)
sigma <- dbnR::calc_sigma(fit)

Calculate the sigma covariance matrix of a Gaussian linear network. This is the C++ backend of the function.

Description

Calculate the sigma covariance matrix of a Gaussian linear network. This is the C++ backend of the function.

Usage

calc_sigma_cpp(fit, order)

Arguments

Value

the covariance matrix

Checks if the vector of names are time formatted to t_0

Description

This will check if the names are properly time formatted in t_0 to be folded into more time slices. A vector is well formatted in t_0 when all of its column names end in '_t_0'.

Usage

check_time0_formatted(obj)

Arguments

Value

TRUE if it is well formatted. FALSE in other case.

Create a matrix with the arcs defined in a causlist object

Description

Create a matrix with the arcs defined in a causlist object

Usage

cl_to_arc_matrix_cpp(cl, ordering, rows)

Arguments

Value

a list with a CharacterVector and a NumericVector

Extracts the coefficients of a DBN

Description

Generic method for "dbn.fit" S3 objects. Calls bnlearn underneath.

Usage

## S3 method for class 'dbn.fit'
coef(object, ...)

Arguments

Value

the coefficients of the network

Creates the blacklist of arcs from a folded data.table

Description

This will create the blacklist of arcs that are not to be learned in the second phase of the dmmhc. This includes arcs backwards in time or inside time-slices.

Usage

create_blacklist(name, size, acc = NULL, slice = 1)

Arguments

Value

the two column matrix with the blacklisted arcs

Create a causal list from a DBN. This is the C++ backend of the function.

Description

Create a causal list from a DBN. This is the C++ backend of the function.

Usage

create_causlist_cpp(cl, net, size, ordering)

Arguments

Value

a list with a CharacterVector and a NumericVector

Create a natural causal list from a DBN. This is the C++ backend of the function.

Description

Create a natural causal list from a DBN. This is the C++ backend of the function.

Usage

create_natcauslist_cpp(cl, net, ordering)

Arguments

Value

the natCauslist equivalent to the DBN

If the names of the nodes have "_t_0" appended at the end, remove it

Description

If the names of the nodes have "_t_0" appended at the end, remove it

Usage

crop_names_cpp(names)

Arguments

Value

the vector with the names cropped

Multiply a Velocity by a constant real number

Description

Multiply a Velocity by a constant real number

Usage

cte_times_vel_cpp(k, vl, abs_op, max_op)

Arguments

Value

a list with the Velocity's new causal list and number of operations

Calculates the degree of a list of nodes

Description

#' Generic method for calculating the degree of a list of nodes in a BN or a DBN. Calls bnlearn's degree underneath. I have to redefine the generic and mask the original for it to work on both bn and dbn objects without the user having to import bnlearn.

Usage

degree(object, Nodes, ...)

Arguments

Value

the degree of the nodes

Calculates the degree of a list of nodes

Description

Generic method for calculating the degree of a list of nodes in a "bn" S3 object. Calls bnlearn's degree underneath.

Usage

## S3 method for class 'bn'
degree(object, Nodes, ...)

Arguments

Value

the degree of the nodes

Calculates the degree of a list of nodes

Description

#' Generic method for calculating the degree of a list of nodes in a "bn.fit" S3 object. Calls bnlearn's degree underneath.

Usage

## S3 method for class 'bn.fit'
degree(object, Nodes, ...)

Arguments

Value

the degree of the nodes

Calculates the degree of a list of nodes

Description

Generic method for calculating the degree of a list of nodes in a "dbn" S3 object. Calls bnlearn's degree underneath. The degree function is specifically coded to look for either the "bn" or the "bn.fit" method inside bnlearn, so I have to cast the "dbn" class into "bn" for it to work.

Usage

## S3 method for class 'dbn'
degree(object, Nodes, ...)

Arguments

Value

the degree of the nodes

Calculates the degree of a list of nodes

Description

#' Generic method for calculating the degree of a list of nodes in a "dbn.fit" S3 object. Calls bnlearn's degree underneath.

Usage

## S3 method for class 'dbn.fit'
degree(object, Nodes, ...)

Arguments

Value

the degree of the nodes

Learns the structure of a markovian n DBN model from data

Description

Learns a gaussian dynamic Bayesian network from a dataset. It allows the creation of markovian n nets rather than only markov 1.

Usage

dmmhc(
  dt,
  size = 2,
  f_dt = NULL,
  blacklist = NULL,
  intra = TRUE,
  blacklist_tr = NULL,
  whitelist = NULL,
  whitelist_tr = NULL,
  ...
)

Arguments

Value

the structure of the net

Gets the ordering of a single time slice in a DBN

Description

This method gets the structure of a DBN, isolates the nodes of a single time slice and then gives a topological ordering of them.

Usage

dynamic_ordering(structure)

Arguments

Value

the ordered nodes of t_0

Performs exact inference forecasting with the GDBN over a dataset

Description

Given a bn.fit object, the size of the net and a data.set, performs exact forecasting over the initial evidence taken from the dataset.

Usage

exact_inference(dt, fit, obj_vars, ini, len, prov_ev)

Arguments

Value

a list with the mu results of the forecast

Performs exact inference smoothing with the GDBN over a dataset

Description

Given a bn.fit object, the size of the net and a dataset, performs exact smoothing over the initial evidence taken from the dataset. Take notice that the smoothing is done backwards in time, as opposed to forecasting.

Usage

exact_inference_backwards(dt, fit, obj_vars, ini, len, prov_ev)

Arguments

Value

a list with the results of the inference backwards

Performs exact inference in a time slice of the dbn

Description

Given a bn.fit object and some variables, performs exact MVN inference over such variables in the net for a given time slice.

Usage

exact_prediction_step(fit, variables, evidence)

Arguments

Value

a list with the predicted mu and sigma

Extends the names of the nodes in t_0 to t_(max-1)

Description

This method extends the names of the nodes to the given maximum and maintains the order of the nodes in each slice, so as to plotting the nodes in all slices relative to their homonyms in the first slice.

Usage

expand_time_nodes(name, acc, max, i)

Arguments

Value

the extended names

Filter the instances in a data.table with different ids in each row

Description

Given an id variable that labels the different instances of a time series inside a dataset, discard the rows that have values from more than 1 id.

Usage

filter_same_cycle(f_dt, size, id_var)

Arguments

Value

the filtered data.table

Examples

dt <- dbnR::motor[201:2500]
dt[, n_sec := rep(seq(46), each = 50)] # I'll create secuences of 50 instances each
f_dt <- dbnR::fold_dt(dt, size = 2)
f_dt[50, .SD, .SDcols = c("n_sec_t_0", "n_sec_t_1")]
f_dt <- dbnR::filter_same_cycle(f_dt, size = 2, id_var = "n_sec")
f_dt[50, .SD, .SDcols = c("n_sec_t_0", "n_sec_t_1")]

Fold a dataset avoiding overlapping of different time series

Description

If the dataset that is going to be folded contains several different time series instances of the same process, folding it could introduce false rows with data from different time series. Given an id variable that labels the different instances of a time series inside a dataset and a desired size, this function folds the dataset and avoids mixing data from different origins in the same instance.

Usage

filtered_fold_dt(dt, size, id_var, clear_id_var = TRUE)

Arguments

Value

the filtered data.table

Examples

dt <- dbnR::motor[201:2500]
dt[, n_sec := rep(seq(46), each = 50)] # I'll create secuences of 50 instances each
f_dt <- dbnR::fold_dt(dt, size = 2)
dim(f_dt)
f_dt <- dbnR::filtered_fold_dt(dt, size = 2, id_var = "n_sec")
dim(f_dt)  # The filtered folded dt has a row less for each independent secuence

Fits a markovian n DBN model

Description

Fits the parameters of the DBN via MLE. The "mu" vector of means and the "sigma" covariance matrix are set as attributes of the dbn.fit object for future exact inference.

Usage

fit_dbn_params(net, f_dt, ...)

Arguments

Value

a "dbn.fit" S3 object with the fitted net

Examples

size = 3
dt_train <- dbnR::motor[200:2500]
net <- learn_dbn_struc(dt_train, size)
f_dt_train <- fold_dt(dt_train, size)
fit <- fit_dbn_params(net, f_dt_train, method = "mle-g")

Extracts the fitted values of a DBN

Description

Generic method for "dbn.fit" S3 objects. Calls bnlearn underneath.

Usage

## S3 method for class 'dbn.fit'
fitted(object, ...)

Arguments

Value

the fitted values of the network

Widens the dataset to take into account the t previous time slices

Description

This function will widen the dataset to put the t previous time slices in each row, so that it can be used to learn temporal arcs in the second phase of the dmmhc.

Usage

fold_dt(dt, size)

Arguments

Value

the extended data.table

Examples

data(motor)
size <-  3
f_dt <- fold_dt(motor, size)

Widens the dataset to take into account the t previous time slices

Description

This will widen the dataset to put the t previous time slices in each row, so that it can be used to learn temporal arcs in the second phase of the dmmhc. Recursive version not exported, the user calls from the handler 'fold_dt'

Usage

fold_dt_rec(dt, n_prev, size, slice = 1)

Arguments

Value

the extended data.table

Performs forecasting with the GDBN over a dataset

Description

Given a dbn.fit object, the size of the net and a folded dataset, performs a forecast over the initial evidence taken from the dataset.

Usage

forecast_ts(
  dt,
  fit,
  size = NULL,
  obj_vars,
  ini = 1,
  len = dim(dt)[1] - ini,
  rep = 1,
  num_p = 50,
  print_res = TRUE,
  plot_res = TRUE,
  mode = "exact",
  prov_ev = NULL
)

Arguments

Value

a list with the original time series values and the results of the forecast

Examples

size = 3
data(motor)
dt_train <- motor[200:900]
dt_val <- motor[901:1000]
obj <- c("pm_t_0")
net <- learn_dbn_struc(dt_train, size)
f_dt_train <- fold_dt(dt_train, size)
f_dt_val <- fold_dt(dt_val, size)
fit <- fit_dbn_params(net, f_dt_train, method = "mle-g")
res <- suppressWarnings(forecast_ts(f_dt_val, fit, 
        obj_vars = obj, len = 10, print_res = FALSE, plot_res = FALSE))

Generate a random DBN and a sampled dataset

Description

This function generates both a random DBN and a dataset that can be used to learn its structure from data. It's intended for experimental use.

Usage

generate_random_network_exp(
  n_vars,
  size,
  min_mu,
  max_mu,
  min_sd,
  max_sd,
  min_coef,
  max_coef,
  seed = NULL
)

Arguments

Value

a list with the original network structure and the sampled dataset

Initialize the nodes vector

Description

Initialize the vector in C++

Usage

init_cl_cpp(n_nodes)

Arguments

Value

a list with the randomly initialized particles

Initialize the particles

Description

Initialize the particles

Usage

init_list_cpp(nodes, size, n_inds)

Arguments

Value

a list with the randomly initialized particles

Create a causality list and initialize it

Description

Create a causality list and initialize it

Usage

initialize_cl_cpp(ordering, size)

Arguments

Value

a causality list

Learns the structure of a markovian n DBN model from data

Description

Learns a gaussian dynamic Bayesian network from a dataset. It allows the creation of markovian n nets rather than only markov 1.

Usage

learn_dbn_struc(dt, size = 2, method = "dmmhc", f_dt = NULL, ...)

Arguments

Value

a "dbn" S3 object with the structure of the network

Examples

data("motor")
net <- learn_dbn_struc(motor, size = 3)

Calculate the log-likelihood of a dynamic Bayesian network

Description

Generic method for calculating the log-likelihood of a "dbn" S3 object given some data. Calls bnlearn's logLik underneath.

Usage

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

Arguments

Value

the log-likelihood score of the network

Calculate the log-likelihood of a dynamic Bayesian network

Description

Generic method for calculating the log-likelihood of a "dbn.fit" S3 object given some data. Calls bnlearn's logLik underneath.

Usage

## S3 method for class 'dbn.fit'
logLik(object, dt, ...)

Arguments

Value

the log-likelihood score of the network

Average the parameters of multiple dbn.fit objects with identical structures

Description

Generic method for "dbn.fit" S3 objects. Calls bnlearn underneath.

Usage

## S3 method for class 'dbn.fit'
mean(x, ...)

Arguments

Value

the averaged parameters

Merges and replicates the arcs in the static BN into all the time-slices in the DBN

Description

This will join the static net and the state transition net by replicating the arcs in the static net in all the time slices.

Usage

merge_nets(net0, netCP1, size, acc = NULL, slice = 1)

Arguments

Value

the merged nets

Multivariate time series dataset on the temperature of an electric motor

Description

Data from several sensors on an electric motor that records different benchmark sessions of measurements at 2 Hz. The dataset is reduced to 3000 instances from the 60th session in order to include it in the package for testing purposes. For the complete dataset, refer to the source.

Usage

data(motor)

Format

An object of class data.table (inherits from data.frame) with 3000 rows and 11 columns.

Source

Kaggle, <https://www.kaggle.com/wkirgsn/electric-motor-temperature>

Performs inference over a multivariate normal distribution

Description

Given some evidence, this function performs inference over a multivariate normal distribution. After converting a Gaussian linear network to its MVN form, this kind of inference can be performed. It's recommended to use predict_dt functions instead unless you need a more flexible inference method.

Usage

mvn_inference(mu, sigma, evidence)

Arguments

Value

a list with the posterior mean and covariance matrix

Examples

size = 3
data(motor)
dt_train <- motor[200:2500]
dt_val <- motor[2501:3000]
obj <- c("pm_t_0")

net <- learn_dbn_struc(dt_train, size)
f_dt_train <- fold_dt(dt_train, size)
f_dt_val <- fold_dt(dt_val, size)
ev <- f_dt_val[1, .SD, .SDcols = obj]
fit <- fit_dbn_params(net, f_dt_train, method = "mle-g")

pred <- mvn_inference(calc_mu(fit), calc_sigma(fit), ev)

R6 class that defines causal lists in the PSO

Description

Constructor of the 'natCauslist' class

Arguments

Details

The causal lists will be the base of the positions and the velocities in the pso part of the algorithm. They will not have the same structure as their binary counterparts, but their class skeleton will serve as a base.

Value

A new 'natCauslist' object

Fields

cl

List of causal units

ordering

String vector defining the order of the nodes in t_0

ordering

String vector defining the order of the nodes without the appended "_t_0"

R6 class that defines a Particle in the PSO algorithm

Description

Constructor of the 'natParticle' class

Evaluate the score of the particle's position

Evaluate the score of the particle's position. Updates the local best if the new one is better.

Update the position of the particle with the velocity

Update the position of the particle given the constants after calculating the new velocity

Arguments

Details

A particle has a Position, a Velocity and a local best

Value

A new 'natParticle' object

The score of the current position

Fields

ps

position of the particle

cl

velocity of the particle

velocity

that takes the particle to the global best

velocity

that takes the particle to the local best

lb

local best score obtained

lb_ps

local best position found

score

bnlearn score function used

R6 class that defines DBNs as vectors of natural numbers

Description

Constructor of the 'natPosition' class

Translate the vector into a DBN network

Uses this object private cl and transforms it into a DBN.

Add a velocity to the position

Given a natVelocity object, add it to the current position.

Return the static node ordering

This function takes as input a dbn and return the node ordering of the variables inside a timeslice. This ordering is needed to understand a position vector.

Translate a DBN into a position vector

This function takes as input a network from a DBN and transforms the structure into a vector of natural numbers if it is a valid DBN. Valid DBNs have only inter-timeslice edges and only allow variables in t_0 to have parents.

Generates a random position

This function takes as input the number of variables, the maximum size and the parameter p and returns a random position with arcs sampled either from the uniform distribution or from a truncated geometric distribution. Much faster than the binary implementation with lists of lists and random bn generation into translation.

Recount the number of arcs in the cl

Arguments

Details

A natPosition represents a single HO-DBN structure with a vector. Its function is to encode the solutions in the PSO framework. Each particle will have a position.

Value

A new 'natPosition' object

a dbn object

the ordering of the nodes in t_0

a random position

the number of arcs

Fields

n_arcs

Number of arcs in the network

max_size

Maximum number of timeslices of the DBN

p

Parameter of the sampling truncated geometric distribution

nodes

Names of the nodes in the network

R6 class that defines the PSO controller

Description

Constructor of the 'natPsoCtrl' class

If the names of the nodes have "_t_0" appended at the end, remove it

Initialize the particles for the algorithm to random positions and velocities.

Arguments

Details

The controller will encapsulate the particles and run the algorithm. This time, it extends the class "PsoCtrl" in the "structure_learning_psoho.R" file, because both controllers are practically the same. The particles, positions and velocities are too different to extend one another though.

Value

A new 'natPsoCtrl' object

the ordering with the names cropped

Learn a DBN structure with a PSO approach

Description

Given a dataset and the desired Markovian order, this function returns a DBN structure ready to be fitted. Original algorithm at https://link.springer.com/chapter/10.1007/978-3-030-86271-8_14

Usage

natPsoho(
  dt,
  size,
  f_dt = NULL,
  n_inds = 50,
  n_it = 50,
  in_cte = 1,
  gb_cte = 0.5,
  lb_cte = 0.5,
  v_probs = c(10, 65, 25),
  r_probs = c(-0.5, 1.5),
  score = "bge",
  p = 0.06,
  cte = TRUE
)

Arguments

Value

A 'dbn' object with the structure of the best network found

R6 class that defines velocities in the PSO

Description

Constructor of the 'natVelocity' class. Only difference with the natCauslist one is that it has a negative cl attribute.

Getter of the abs_op attribute.

return the number of operations that the velocity performs

Setter of the abs_op attribute. Intended for inside use only. This should be a 'protected' function in Java-like OOP, but there's no such thing in R6. This function should not be used from outside the package.

Randomizes the Velocity's directions.

Given two positions, returns the velocity that gets the first position to the other one.

Add both velocities directions

Multiply the Velocity by a constant real number

This function multiplies the Velocity by a constant real number. It is non deterministic by definition. When calculating k*|V|, the result will be floored and bounded to the set [-max_op, max_op], where max_op is the maximum number of arcs that can be present in the network.

Arguments

Details

The velocities will be defined as two natural vectors where each element in them represents the arcs from a temporal family of nodes to a receiving node. 1-bits in the binary representation of this number represent arc additions/deletions

Value

A new 'natVelocity' object

the natVelocity that gets the ps1 to ps2

Fields

abs_op

Total number of operations 1 or -1 in the velocity

max_size

Maximum number of timeslices of the DBN

cl_neg

Negative part of the velocity

Create a matrix with the arcs defined in a causlist object

Description

Create a matrix with the arcs defined in a causlist object

Usage

nat_cl_to_arc_matrix_cpp(cl, ordering, rows)

Arguments

Value

a StringMatrix with the parent nodes and the children nodes

Multiply a Velocity by a constant real number

Description

Multiply a Velocity by a constant real number

Usage

nat_cte_times_vel_cpp(k, vl, vl_neg, abs_op, max_size)

Arguments

Value

the new total number of operations

Subtracts two natPositions to obtain the natVelocity that transforms ps1 into ps2

Description

Subtracts two natPositions to obtain the natVelocity that transforms ps1 into ps2

Usage

nat_pos_minus_pos_cpp(ps1, ps2, vl, vl_neg)

Arguments

Value

the velocity's causal lists by reference and the number of operations by return

Add a velocity to a position

Description

Add a velocity to a position

Usage

nat_pos_plus_vel_cpp(cl, vl, vl_neg, n_arcs)

Arguments

Value

the new position by reference and the new number of arcs by return

Adds two natVelocities

Description

Adds two natVelocities represented as two numeric vectors: one with the positive part and one with the negative part. Adding them is a process that does a bitwise 'or' with both the positive and negative parts of the two velocities, adjusts the new abs_op, removes duplicated arcs in the final velocity by using a bitwise 'xor' with both parts and adjusts the final abs_op. The results are returned via modifying the original vl1 and vl1_neg by reference and returning the final abs_op normally. I can't have an integer edited by reference because it automatically gets casted and cannot be used to return values.

Usage

nat_vel_plus_vel_cpp(vl1, vl1_neg, vl2, vl2_neg, abs_op1, abs_op2)

Arguments

Value

the total number of resulting operations

Defines a level for every node in the net

Description

Calculates the levels in which the nodes will be distributed when plotting the structure. This level is defined by their parent nodes: a node with no parents will always be in the level 0. Subsequently, the level of a node will be one more of the maximum level of his parents.

Usage

node_levels(net, order, lvl = 1, acc = NULL)

Arguments

Value

a matrix with the names of the nodes in the first row and their level on the second

Returns a list with the names of the nodes of a BN or a DBN

Description

Generic method for obtaining the names of the nodes in a BN or a DBN. Calls bnlearn's nodes underneath. I have to redefine the generic and mask the original for it to work on both bn and dbn objects without the user having to import bnlearn.

Usage

nodes(object, ...)

Arguments

Value

the names of the nodes

Relabel the names of the nodes of a BN or a DBN

Description

Generic method for renaming the nodes in a BN or a DBN. Calls bnlearn's nodes<- underneath. I have to redefine the generic and mask the original for it to work on both bn and dbn objects without the user having to import bnlearn.

Usage

nodes(object) <- value

Arguments

Value

the modified object

Relabel the names of the nodes of a BN or a DBN

Description

Generic method for renaming the nodes in a "bn" S3 object. Calls bnlearn's nodes<- underneath.

Usage

## S3 replacement method for class 'bn'
nodes(object) <- value

Arguments

Value

the modified object

Relabel the names of the nodes of a BN or a DBN

Description

Generic method for renaming the nodes in a "bn.fit" S3 object. Calls bnlearn's nodes<- underneath.

Usage

## S3 replacement method for class 'bn.fit'
nodes(object) <- value

Arguments

Value

the modified object

Relabel the names of the nodes of a BN or a DBN

Description

Generic method for renaming the nodes in a "bn" S3 object. Calls bnlearn's nodes<- underneath.

Usage

## S3 replacement method for class 'dbn'
nodes(object) <- value

Arguments

Value

the modified object

Relabel the names of the nodes of a BN or a DBN

Description

Generic method for renaming the nodes in a "bn" S3 object. Calls bnlearn's nodes<- underneath.

Usage

## S3 replacement method for class 'dbn.fit'
nodes(object) <- value

Arguments

Value

the modified object

Returns a list with the names of the nodes of a BN or a DBN

Description

Generic method for obtaining the names of the nodes in "bn" S3 object. Calls bnlearn's nodes underneath.

Usage

## S3 method for class 'bn'
nodes(object, ...)

Arguments

Value

the names of the nodes

Returns a list with the names of the nodes of a BN or a DBN

Description

Generic method for obtaining the names of the nodes in "bn.fit" S3 object. Calls bnlearn's nodes underneath.

Usage

## S3 method for class 'bn.fit'
nodes(object, ...)

Arguments

Value

the names of the nodes

Returns a list with the names of the nodes of a BN or a DBN

Description

Generic method for obtaining the names of the nodes in "dbn" S3 object. Calls bnlearn's nodes underneath. The nodes function is specifically coded to look for either the "bn" or the "bn.fit" method inside bnlearn, so I have to cast the "dbn" class into "bn" for it to work.

Usage

## S3 method for class 'dbn'
nodes(object, ...)

Arguments

Value

the names of the nodes

Returns a list with the names of the nodes of a BN or a DBN

Description

Generic method for obtaining the names of the nodes in "dbn.fit" S3 object. Calls bnlearn's nodes underneath.

Usage

## S3 method for class 'dbn.fit'
nodes(object, ...)

Arguments

Value

the names of the nodes

Generates the names of the nodes in t_0 and in all the network

Description

Given the names of the desired variables, this function generates the names of the variables in a DBN without needing a previous dataset. It's just a wrapper around the 'fold_dt' function.

Usage

nodes_gen_exp(ordering, size)

Arguments

Value

a dictionary with the variable names in t_0 and in all other time slices

One hot encoder for natural numbers without the 0.

Description

Given a natural number, return the natural number equivalent to its one-hot encoding. Examples: 3 -> 100 -> 4, 5 -> 10000 -> 16

Usage

one_hot(nat)

Arguments

Value

the converted number

One-hot encoder for natural numbers without the 0

Description

Given a natural number, return the natural number equivalent to its one-hot encoding. Instead of pow, the '<<' operator will be used. Examples: 3 -> 100 -> 4, 5 -> 10000 -> 16

Usage

one_hot_cpp(nat)

Arguments

Value

the converted number

Generates the names of n variables.

Description

Given the total number of variables, this function generates a character vector with variables named "Xi", where i is a number in the interval [0,n-1]

Usage

ordering_gen_exp(n)

Arguments

Value

a character vector with the variable names

Plots a dynamic Bayesian network

Description

Generic method for plotting the "dbn" S3 objects. Calls plot_dynamic_network underneath.

Usage

## S3 method for class 'dbn'
plot(x, ...)

Arguments

Plots a fitted dynamic Bayesian network

Description

Generic method for plotting the "dbn.fit" S3 objects. Calls plot_dynamic_network underneath.

Usage

## S3 method for class 'dbn.fit'
plot(x, ...)

Arguments

Plots a dynamic Bayesian network in a hierarchical way

Description

To plot the DBN, this method first computes a hierarchical structure for a time slice and replicates it for each slice. Then, it calculates the relative position of each node with respect to his equivalent in the first slice. The result is a net where each time slice is ordered and separated from one another, where the leftmost slice is the oldest and the rightmost represents the present time. This function is also called by the generic plot function of "dbn" and "dbn.fit" S3 objects.

Usage

plot_dynamic_network(
  structure,
  offset = 200,
  subset_nodes = NULL,
  reverse = FALSE
)

Arguments

Value

the visualization of the DBN

Examples

size = 3
dt_train <- dbnR::motor[200:2500]
net <- learn_dbn_struc(dt_train, size)
plot_dynamic_network(net)

Plots a Bayesian network in a hierarchical way

Description

This function calculates the levels of each node and then plots them in a hierarchical layout in visNetwork. Can be used in place of the generic plot function offered by bnlearn for "bn" and "bn.fit" S3 objects.

Usage

plot_static_network(structure)

Arguments

Examples

dt_train <- dbnR::motor[200:2500]
net <- bnlearn::mmhc(dt_train)
plot_static_network(net)
fit <- bnlearn::bn.fit(net, dt_train, method = "mle-g")
plot_static_network(fit) # Works for both the structure and the fitted net

Subtracts two Positions to obtain the Velocity that transforms one into the other

Description

Subtracts two Positions to obtain the Velocity that transforms one into the other

Usage

pos_minus_pos_cpp(cl, ps, vl)

Arguments

Value

a list with the Velocity's causal list and the number of operations

Add a velocity to a position

Description

Add a velocity to a position

Usage

pos_plus_vel_cpp(cl, vl, n_arcs)

Arguments

Value

a list with the modified position and the new number of arcs

Performs inference in every row of a dataset with a DBN

Description

Generic method for predicting a dataset with a "dbn.fit" S3 objects. Calls predict_dt underneath.

Usage

## S3 method for class 'dbn.fit'
predict(object, ...)

Arguments

Value

a data.table with the prediction results

Performs inference over a fitted GBN

Description

Performs inference over a Gaussian BN. It's thought to be used in a map for a data.table, to use as evidence each separate row. If not specifically needed, it's recommended to use the function predict_dt instead. This function is deprecated and will be removed in a future version.

Usage

predict_bn(fit, evidence)

Arguments

Value

a data.table with the predictions

Examples

size = 3
data(motor)
dt_train <- motor[200:2500]
dt_val <- motor[2501:3000]
net <- learn_dbn_struc(dt_train, size)
f_dt_train <- fold_dt(dt_train, size)
f_dt_val <- fold_dt(dt_val, size)
fit <- fit_dbn_params(net, f_dt_train, method = "mle-g")
res <- f_dt_val[, predict_bn(fit, .SD), .SDcols = c("pm_t_0", "coolant_t_0"), by = 1:nrow(f_dt_val)]

Performs inference over a test dataset with a GBN

Description

This function performs inference over each row of a folded data.table, plots the results and gives metrics of the accuracy of the predictions. Given that only a single row is predicted, the horizon of the prediction is at most 1. This function is also called by the generic predict method for "dbn.fit" objects. For long term forecasting, please refer to the forecast_ts function.

Usage

predict_dt(fit, dt, obj_nodes, verbose = T, look_ahead = F)

Arguments

Value

a data.table with the prediction results for each row

Examples

size = 3
data(motor)
dt_train <- motor[200:900]
dt_val <- motor[901:1000]

# With a DBN
obj <- c("pm_t_0")
net <- learn_dbn_struc(dt_train, size)
f_dt_train <- fold_dt(dt_train, size)
f_dt_val <- fold_dt(dt_val, size)
fit <- fit_dbn_params(net, f_dt_train, method = "mle-g")
res <- suppressWarnings(predict_dt(fit, f_dt_val, obj_nodes = obj, verbose = FALSE))

# With a Gaussian BN directly from bnlearn
obj <- c("pm")
net <- bnlearn::mmhc(dt_train)
fit <- bnlearn::bn.fit(net, dt_train, method = "mle-g")
res <- suppressWarnings(predict_dt(fit, dt_val, obj_nodes = obj, verbose = FALSE))

Print method for "dbn" objects

Description

Generic print method for "dbn" S3 objects. Calls bnlearn's print underneath

Usage

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

Arguments

Print method for "dbn.fit" objects

Description

Generic print method for "dbn.fit" S3 objects. Calls bnlearn's print underneath

Usage

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

Arguments

Learn a DBN structure with a PSO approach

Description

Given a dataset and the desired Markovian order, this function returns a DBN structure ready to be fitted. It requires a folded dataset. Original algorithm at https://doi.org/10.1109/BRC.2014.6880957

Usage

psoho(
  dt,
  size,
  f_dt = NULL,
  n_inds = 50,
  n_it = 50,
  in_cte = 1,
  gb_cte = 0.5,
  lb_cte = 0.5,
  v_probs = c(10, 65, 25),
  r_probs = c(-0.5, 1.5),
  score = "bge",
  cte = TRUE
)

Arguments

Value

A 'dbn' object with the structure of the best network found

Randomize a velocity with the given probabilities

Description

Randomize a velocity with the given probabilities

Usage

randomize_vl_cpp(vl, probs)

Arguments

Value

a velocity list with randomized values

Simulates random samples from a fitted DBN

Description

Generic method for "dbn.fit" S3 objects. Calls bnlearn's rbn underneath.

Usage

rbn.dbn.fit(x, n, ...)

Arguments

Value

the sampled dataset

Experimental function that recounts the number of arcs in the position

Description

Experimental function that recounts the number of arcs in the position

Usage

recount_arcs_exp(ps)

Arguments

Value

the number of arcs

Reduce the frequency of the time series data in a data.table

Description

In a time series dataset, there is a time difference between one row and the next one. This function reduces the number of rows from its current frequency to the desired one by averaging batches of rows. Instead of the frequency in Hz, the number of seconds between rows is asked (Hz = 1/s).

Usage

reduce_freq(dt, obj_freq, curr_freq, id_var = NULL)

Arguments

Value

the data.table with the desired frequency

Examples

# Let's assume that the dataset has a frequency of 4Hz, 0.25 seconds between rows
dt <- dbnR::motor
dim(dt)
# Let's change the frequency to 2Hz, 0.5 seconds between rows
dt <- reduce_freq(dt, obj_freq = 0.5, curr_freq = 0.2)
dim(dt)

Return a list of nodes with the time slice appended up to the desired size of the network

Description

Return a list of nodes with the time slice appended up to the desired size of the network

Usage

rename_nodes_cpp(nodes, size)

Arguments

Value

a list with the renamed nodes in each timeslice

Returns the residuals from fitting a DBN

Description

Generic method for "dbn.fit" S3 objects. Calls bnlearn underneath.

Usage

## S3 method for class 'dbn.fit'
residuals(object, ...)

Arguments

Value

the residuals of fitting the network

Computes the score of a BN or a DBN

Description

Generic method for computing the score of a BN or a DBN. Calls bnlearn's nodes underneath. I have to redefine the generic and mask the original for it to work on both bn and dbn objects without the user having to import bnlearn.

Usage

score(object, ...)

Arguments

Value

the score of the network

Computes the score of a BN or a DBN

Description

Generic method for computing the score of a "bn" S3 object. Calls bnlearn's score underneath.

Usage

## S3 method for class 'bn'
score(object, ...)

Arguments

Value

the score of the network

Computes the score of a BN or a DBN

Description

Generic method for computing the score of a "dbn" S3 object. Calls bnlearn's score underneath. The nodes function is specifically coded to look for either the "bn" or the "bn.fit" method inside bnlearn, so I have to cast the "dbn" class into "bn" for it to work.

Usage

## S3 method for class 'dbn'
score(object, ...)

Arguments

Value

the score of the network

Move the window of values backwards in a folded dataset row

Description

This function moves the values in t_0, t_1, ..., t_n-1 in a folded dataset row to t_1, t_2, ..., t_n. All the variables in t_0 will be inputed with NAs and the obtained row can be used to forecast up to any desired point.

Usage

shift_values(f_dt, row)

Arguments

Value

a one row data.table the shifted values

Examples

dt <- dbnR::motor
f_dt <- dbnR::fold_dt(dt, size = 2)
s_row <- dbnR::shift_values(f_dt, row = 500)

Returns the standard deviation of the residuals from fitting a DBN

Description

Generic method for "dbn.fit" S3 objects. Calls bnlearn underneath.

Usage

## S3 method for class 'dbn.fit'
sigma(object, ...)

Arguments

Value

the standard deviation residuals of fitting the network

Performs smoothing with the GDBN over a dataset

Description

Given a dbn.fit object, the size of the net and a folded dataset, performs a smoothing of a trajectory. Smoothing is the opposite of forecasting: given a starting point, predict backwards in time to obtain the time series that generated that point.

Usage

smooth_ts(
  dt,
  fit,
  size = NULL,
  obj_vars,
  ini = dim(dt)[1],
  len = ini - 1,
  print_res = TRUE,
  plot_res = TRUE,
  prov_ev = NULL
)

Arguments

Value

a list with the original values and the results of the smoothing

Examples

size = 3
data(motor)
dt_train <- motor[200:900]
dt_val <- motor[901:1000]
obj <- c("pm_t_2")
net <- learn_dbn_struc(dt_train, size)
f_dt_train <- fold_dt(dt_train, size)
f_dt_val <- fold_dt(dt_val, size)
fit <- fit_dbn_params(net, f_dt_train, method = "mle-g")
res <- suppressWarnings(smooth_ts(f_dt_val, fit, 
        obj_vars = obj, len = 10, print_res = FALSE, plot_res = FALSE))

Renames the columns in a data.table so that they end in '_t_0'

Description

This will rename the columns in a data.table so that they end in '_t_0', which will be needed when folding the data.table. If any of the columns already ends in '_t_0', a warning will be issued and no further operation will be done. There is no need to use this function to learn a DBN unless some operation with the variable names wants to be done prior to folding a dataset.

Usage

time_rename(dt)

Arguments

Value

the renamed data.table

Examples

data("motor")
dt <- time_rename(motor)

Geometric distribution sampler truncated to a maximum

Description

A geometric distribution sampler with probability 'p' restricted to values inside [1, max]. Because of this restriction, very low values of 'p' coupled with low 'max' return increasingly uniform populations in the interval [1, max].

Usage

trunc_geom(p, max)

Arguments

Value

the sampled value

Add two Velocities

Description

Add two Velocities

Usage

vel_plus_vel_cpp(vl1, vl2, abs_op)

Arguments

Value

a list with the Velocity's causal list and the number of operations