LogicRegr {EMJMCMC}

A wrapper for running the Bayesian logic regression based inference in a easy to use way

Description

A wrapper for running the Bayesian logic regression based inference in a easy to use way

Usage

LogicRegr(
  formula,
  data,
  family = "Gaussian",
  prior = "J",
  report.level = 0.5,
  d = 20,
  cmax = 5,
  kmax = 20,
  p.and = 0.9,
  p.not = 0.05,
  p.surv = 0.1,
  ncores = -1,
  n.mods = 1000,
  print.freq = 1000L,
  advanced = list(presearch = TRUE, locstop = FALSE, estimator =
    estimate.logic.bern.tCCH, estimator.args = list(data = data.example, n = 1000, m =
    50, r = 1), recalc_margin = 250, save.beta = FALSE, interact = TRUE, relations =
    c("", "lgx2", "cos", "sigmoid", "tanh", "atan", "erf"), relations.prob = c(0.4, 0, 0,
    0, 0, 0, 0), interact.param = list(allow_offsprings = 1, mutation_rate = 300,
    last.mutation = 5000, max.tree.size = 1, Nvars.max = 100, p.allow.replace = 0.9,
    p.allow.tree = 0.2, p.nor = 0.2, p.and = 1), 
     n.models = 10000, unique = TRUE,
    max.cpu = ncores, max.cpu.glob = ncores, create.table = FALSE, create.hash = TRUE,
    pseudo.paral = TRUE, burn.in = 50, outgraphs = FALSE, print.freq = print.freq,
    advanced.param = list(max.N.glob = as.integer(10), min.N.glob = as.integer(5), max.N
    = as.integer(3), min.N = as.integer(1), printable = FALSE))
)

Arguments

formula

a formula object for the model to be addressed

data

a data frame object containing variables and observations corresponding to the formula used

family

a string taking values of either "Gaussian" or "Bernoulli" corresponding to the linear or logistic Bayesian logic regression contexts

prior

character values "J" or "G" corresponding either to Jeffey's or robust g prior

report.level

a numeric value in (0,1) specifying the threshold for detections based on the marginal inclusion probabilities

d

population size for the GMJMCMC algorithm

cmax

the maximal allowed depth of logical expressions to be considered

kmax

the maximal number of logical expressions per model

p.and

probability of AND parameter of GMJMCMC algorithm

p.not

probability of applying logical NOT in GMJMCMC algorithm

p.surv

minimal survival probabilities for the features to be allowed to enter the next population

ncores

the maximal number of cores (and GMJMCMC threads) to be addressed in the analysis

n.mods

the number of the best models in the thread to calculate marginal inclusion probabilities

print.freq

printing frequency of the intermediate results

advanced

should only be addressed by experienced users to tune advanced parameters of GMJMCMC, advanced corresponds to the vector of tuning parameters of runemjmcmc function

Value

a list of

feat.stat

detected logical expressions and their marginal inclusion probabilities

predictions

NULL currently, since LogrRegr function is not designed for predictions at the moment, which is still possible in its expert mother function pinferunemjmcmc

allposteriors

all visited by GMJMCMC logical expressions and their marginal inclusion probabilities

threads.stats

a vector of detailed outputs of individual ncores threads of GMJMCMC run

See Also

runemjmcmc pinferunemjmcmc

Examples

set.seed(040590)
X1 <- as.data.frame(
  array(
    data = rbinom(n = 50 * 1000, size = 1,
    prob = runif(n = 50 * 1000, 0, 1)), dim = c(1000, 50)
  )
)
Y1 <- rnorm(
  n = 1000,
  mean = 1 + 0.7 * (X1$V1 * X1$V4) + 0.8896846 * (X1$V8 * X1$V11) + 1.434573 * (X1$V5 * X1$V9),
  sd = 1
)
X1$Y1 <- Y1

# specify the initial formula
formula1 <- as.formula(
  paste(colnames(X1)[51], "~ 1 +", paste0(colnames(X1)[-c(51)], collapse = "+"))
)
data.example <- as.data.frame(X1)


# run the inference with robust g prior
n_cores <- 1L


  res4G <- LogicRegr(
    formula = formula1, data = data.example, family = "Gaussian", prior = "G",
    report.level = 0.5, d = 15, cmax = 2, kmax = 15, p.and = 0.9, p.not = 0.01,
    p.surv = 0.2, ncores = n_cores
  )
  print(res4G$feat.stat)

  # run the inference with Jeffrey's prior
  res4J <- LogicRegr(
    formula = formula1, data = data.example, family = "Gaussian", prior = "J",
    report.level = 0.5, d = 15, cmax = 2, kmax = 15, p.and = 0.9, p.not = 0.01,
    p.surv = 0.2, ncores = n_cores
  )
  print(res4J$feat.stat)