In USCbiostats/causnet: What the Package Does (One Line, Title Case)

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

library(causnet)
set.seed(1234)

causnet() can be run without setting the argument score_fun. In that case causnet:::score_bic_lm is used as the default.

mydata <- simdat(n_var = 5)
causnet(data = mydata, max_parents = 3)

The Default scoring function is score_bic_lm() which is formulated as follows:

score_bic_lm <- function(y, x, mydat) {
  y_nm <- colnames(mydat)[y]
  if (is.element(x[1], seq_len(ncol(mydat)))) {
    x_nms <- colnames(mydat)[x]
  } else {
    x_nms <- "1"
  }
  fit <- lm(paste0(y_nm, " ~ ", paste(x_nms, collapse = " + ")), data = mydat)
  bic <- - (1 / 2) * BIC(fit)
  return(bic)
}

If you want to specify your own scoring you need to create a function that takes 3 arguments and returns a single numeric. Bigger values are better score for node y and parents x.

The first argument is passed in will be a single integer denoting the index of the response variable in the data. The second argument is a vector of integer denoting the dependant variables. Note that this can be NA if there are not dependant variables. And the last argument is the data.

Suppose we want a score function that gives back the average correlation between y and x. This score function can be written as follows. Notice that we are handling the special case where x = NA.

mean_cor_score <- function(y, x, mydat) {
  if (any(is.na(x))) return(0)

  mean(cor(mydat[x], mydat[y]))
}

Now we simply pass our custom scoring function to score_fun and run

causnet(data = mydata, max_parents = 3, score_fun = mean_cor_score)

USCbiostats/causnet documentation built on July 15, 2020, 10:41 a.m.