inst/examples/bayesian_swissmetro/plot_b07_discrete_mixture.R

In rbiogeme: Interface to 'Biogeme'

#!/usr/bin/env Rscript

# b07. Bayesian discrete mixture of two logit models.
#
# Class 1 has no time coefficient. Class 2 estimates a common time
# coefficient. The class probabilities are normalized symbolically in R and
# the complete mixture log-likelihood is compiled to native Biogeme.

library(rbiogeme)

# prepare_swissmetro_example() is defined in ../swissmetro/example_utils.R.
# It makes this script independent of the current working directory and
# documents the --data, --python, and --output command-line interface.
script_path <- commandArgs(trailingOnly = FALSE)
script_path <- sub("^--file=", "", script_path[startsWith(script_path, "--file=")][[1L]])
source(file.path(dirname(normalizePath(script_path)), "..", "swissmetro", "example_utils.R"))

build_b07_discrete_mixture_model <- function(database) {
  # Parameter definitions preserve native names, starts, bounds, and status.
  asc_car <- biogeme_beta("asc_car", start = 0)
  asc_train <- biogeme_beta("asc_train", start = 0)
  asc_sm <- biogeme_beta("asc_sm", start = 0, fixed = TRUE)
  b_time <- biogeme_beta("b_time", start = 0)
  b_cost <- biogeme_beta("b_cost", start = 0)
  prob_class1 <- biogeme_beta(
    "prob_class1",
    start = 0.5,
    lower = 0,
    upper = 1
  )
  prob_class2 <- 1 - prob_class1

  availability <- list(
    `1` = variable("TRAIN_AV_SP"),
    `2` = variable("SM_AV"),
    `3` = variable("CAR_AV_SP")
  )
  choice <- variable("CHOICE")

  # Class 1 utilities omit travel time, exactly as in the native example.
  utilities_class_1 <- list(
    `1` = asc_train + b_cost * variable("TRAIN_COST_SCALED"),
    `2` = asc_sm + b_cost * variable("SM_COST_SCALED"),
    `3` = asc_car + b_cost * variable("CAR_CO_SCALED")
  )
  utilities_class_2 <- list(
    `1` = asc_train + b_time * variable("TRAIN_TT_SCALED") +
      b_cost * variable("TRAIN_COST_SCALED"),
    `2` = asc_sm + b_time * variable("SM_TT_SCALED") +
      b_cost * variable("SM_COST_SCALED"),
    `3` = asc_car + b_time * variable("CAR_TT_SCALED") +
      b_cost * variable("CAR_CO_SCALED")
  )

  # logit_probability() is the native probability expression. The weighted
  # sum is then logged symbolically; Biogeme evaluates the whole tree.
  probability_class_1 <- logit_probability(
    utilities_class_1,
    availability,
    alternative = choice
  )
  probability_class_2 <- logit_probability(
    utilities_class_2,
    availability,
    alternative = choice
  )
  log_probability <- log(
    prob_class1 * probability_class_1 + prob_class2 * probability_class_2
  )

  biogeme_model(
    database = database,
    formula = log_probability,
    control = biogeme_control(
    output_directory = prepared$output,
      model_name = "b07_discrete_mixture",
      generate_html = TRUE,
      generate_yaml = TRUE,
      generate_netcdf = TRUE
    )
  )
}

prepared <- prepare_swissmetro_example(
  commandArgs(trailingOnly = TRUE),
  default_model = "b07_discrete_mixture"
)

# Do not allow a previous Bayesian result or iteration file to be reused.
unlink(file.path(prepared$output, c(
  "b07_discrete_mixture.yaml",
  "b07_discrete_mixture.nc",
  "b07_discrete_mixture.html",
  "__b07_discrete_mixture.iter"
)), force = TRUE)

database <- swissmetro_data(prepared$data)
model <- build_b07_discrete_mixture_model(database)
fit <- bayesian_estimate(model, model_name = "b07_discrete_mixture", control = model$control)

print(summary(fit))
print(coef(fit))
print(bayesian_stored_variables(fit))
invisible(fit)