forward_greedy: A forward greedy search function
In kelvinyangli/mbmml: Learn Markov blankets using Minimum Message Length

This is a forward greedy seearch function that can be used incorprate with an objective function mml cpt, logit, or naive bayes (adaptive code) to discovery Markov blanket candidates. It is a greedy search, so the function will stop if there is no better option to add into the current Markov blanket.

1
2
3

forward_greedy(data, arities, vars, sampleSize, target, model, sigma = 3,
  dataNumeric = NULL, varCnt = NULL, prior = "uniform", dag = NULL,
  alpha = 1, statingPara = FALSE, debug = FALSE)

`data`	A dataset whost variables are in numeric/integer format. Any categorical variables must be converted into numeric/integer first.
`arities`	A vector of variable arities in data, in the same order as the column names of data.
`vars`	A vector of all variables in data, in the same order as the column names of data.
`sampleSize`	The sample size. That is, the number of rows of data.
`target`	The target node, whose Markov blanket we are interested in.
`model`	The options are cpt, logit (binary only), naive bayes and random models.
`sigma`	The standard derivation of the assumed Gaussian distribution for parameter prior. The default value is 3 as suggested by the original paper.
`dataNumeric`	This parameter is for mml_logit. The numeric format of the given data set. Variable values start from 0.
`varCnt`	This parameter is for mml_cpt. As explained by argument name. It is obtained by getting the detailed information of the given data using the function count_occurance().
`prior`	A character parameter with options "uniform", "tom" and "bayes" indicate the uniform prior (default), TOM (totally ordered model) and Bayesian prior when averaging the message lengths for random structures. The Bayesian prior starts with the uniform prior then calculates the posteriors and use them as priors for the next step.
`dag`	The true DAG.
`alpha`	A vector of concentration parameters for a Dirichlet distribution. Range is from zeor to positive infinity, length is equal to the arity of the target variable.
`statingPara`	Default is FALSE. If TRUE, then MML estimate of the parameters are also stated with extra 0.5log(pi*e/6) per parameter.
`debug`	A boolean argument to show the detailed Markov blanket inclusion steps based on each mml score.