R/generate-random_model_uni.R

In BKT: Bayesian Knowledge Tracing Model

random_model_uni <- function(num_resources = 1, num_subparts = 1,
                             trans_prior = NULL, given_notknow_prior = NULL,
                             given_know_prior = NULL, pi_0_prior = NULL) {
  # dismiss handle
  if (is.null(trans_prior)) {
    trans_prior <- t(array(c(20, 4, 1, 20), dim = c(2, 2)))
    trans_prior <- reshape_python(rep(trans_prior, num_resources), dim = c(num_resources, 2, 2))
  }
  if (is.null(given_notknow_prior)) {
    given_notknow_prior <- array(rep(c(5, 0.5), num_subparts), dim = c(2, num_subparts))
  }
  if (is.null(given_know_prior)) {
    given_know_prior <- array(rep(c(0.5, 5), num_subparts), dim = c(2, num_subparts))
  }
  if (is.null(pi_0_prior)) {
    pi_0_prior <- matrix(c(100, 1), nrow = 2, byrow = TRUE)
  }

  # model parameter generate
  As <- dirrnd(trans_prior)
  given_notknow <- dirrnd(given_notknow_prior)
  given_know <- dirrnd(given_know_prior)
  pi_0 <- dirrnd(pi_0_prior)

  given_notknow_reshaped <- t(matrix(given_notknow, nrow = 2, ncol = num_subparts))
  given_know_reshaped <- t(matrix(given_know, nrow = 2, ncol = num_subparts))
  emissions <- array(c(given_notknow_reshaped, given_know_reshaped),
    dim = c(num_subparts, 2, 2)
  )
  emissions <- aperm(emissions, c(1, 3, 2))

  modelstruct <- list()
  modelstruct$prior <- runif(1)

  As[, 2, 1] <- runif(num_resources) * 0.40
  As[, 2, 2] <- 1 - As[, 2, 1]
  As[, 1, 2] <- 0
  As[, 1, 1] <- 1

  modelstruct$learns <- As[, 2, 1]
  modelstruct$forgets <- As[, 1, 2]
  given_notknow[2, ] <- runif(num_subparts) * 0.40
  modelstruct$guesses <- given_notknow[2, ]
  given_know[1, ] <- runif(num_subparts) * 0.30
  modelstruct$slips <- given_know[1, ]

  modelstruct$As <- As
  modelstruct$emissions <- emissions
  modelstruct$pi_0 <- pi_0

  return(modelstruct)
}