R/lencode_bayes.R

In embed: Extra Recipes for Encoding Predictors

#' Supervised Factor Conversions into Linear Functions using Bayesian Likelihood
#' Encodings
#'
#' `step_lencode_bayes()` creates a *specification* of a recipe step that will
#' convert a nominal (i.e. factor) predictor into a single set of scores derived
#' from a generalized linear model estimated using Bayesian analysis.
#'
#' @param recipe A recipe object. The step will be added to the sequence of
#'   operations for this recipe.
#' @param ... One or more selector functions to choose variables. For
#'   `step_lencode_bayes`, this indicates the variables to be encoded into a
#'   numeric format. See [recipes::selections()] for more details. For the
#'   `tidy` method, these are not currently used.
#' @param role Not used by this step since no new variables are created.
#' @param outcome A call to `vars` to specify which variable is used as the
#'   outcome in the generalized linear model. Only numeric and two-level factors
#'   are currently supported.
#' @param options A list of options to pass to [rstanarm::stan_glmer()].
#' @param verbose A logical to control the default printing by
#'   [rstanarm::stan_glmer()].
#' @param mapping A list of tibble results that define the encoding. This is
#'   `NULL` until the step is trained by [recipes::prep()].
#' @param skip A logical. Should the step be skipped when the recipe is baked by
#'   [recipes::bake()]? While all operations are baked when [recipes::prep()] is
#'   run, some operations may not be able to be conducted on new data (e.g.
#'   processing the outcome variable(s)). Care should be taken when using `skip
#'   = TRUE` as it may affect the computations for subsequent operations
#' @param trained A logical to indicate if the quantities for preprocessing have
#'   been estimated.
#' @param id A character string that is unique to this step to identify it.
#' @return An updated version of `recipe` with the new step added to the
#'   sequence of existing steps (if any). For the `tidy` method, a tibble with
#'   columns `terms` (the selectors or variables for encoding), `level` (the
#'   factor levels), and `value` (the encodings).
#' @keywords datagen
#' @concept preprocessing encoding
#' @details
#'
#' For each factor predictor, a generalized linear model is fit to the outcome
#' and the coefficients are returned as the encoding. These coefficients are on
#' the linear predictor scale so, for factor outcomes, they are in log-odds
#' units. The coefficients are created using a no intercept model and, when two
#' factor outcomes are used, the log-odds reflect the event of interest being
#' the _first_ level of the factor.
#'
#' For novel levels, a slightly timmed average of the coefficients is returned.
#'
#' A hierarchical generalized linear model is fit using [rstanarm::stan_glmer()]
#' and no intercept via
#'
#' ```
#'   stan_glmer(outcome ~ (1 | predictor), data = data, ...)
#' ```
#'
#' where the `...` include the `family` argument (automatically set by the step,
#' unless passed in by `options`) as well as any arguments given to the
#' `options` argument to the step. Relevant options include `chains`, `iter`,
#' `cores`, and arguments for the priors (see the links in the References
#' below). `prior_intercept` is the argument that has the most effect on the
#' amount of shrinkage.
#'
#' # Tidying
#'
#' When you [`tidy()`][tidy.recipe()] this step, a tibble is retruned with
#' columns `level`, `value`, `terms`, and `id`:
#' 
#' \describe{
#'   \item{level}{character, the factor levels}
#'   \item{value}{numeric, the encoding}
#'   \item{terms}{character, the selectors or variables selected}
#'   \item{id}{character, id of this step}
#' }
#' 
#' @template case-weights-supervised
#'
#' @references
#'
#' Micci-Barreca D (2001) "A preprocessing scheme for high-cardinality
#' categorical attributes in classification and prediction problems," ACM SIGKDD
#' Explorations Newsletter, 3(1), 27-32.
#'
#' Zumel N and Mount J (2017) "vtreat: a data.frame Processor for Predictive
#' Modeling," arXiv:1611.09477
#'
#' "Hierarchical Partial Pooling for Repeated Binary Trials"
#' \url{https://CRAN.R-project.org/package=rstanarm/vignettes/pooling.html}
#'
#' "Prior Distributions for `rstanarm` Models"
#' \url{http://mc-stan.org/rstanarm/reference/priors.html}
#'
#' "Estimating Generalized (Non-)Linear Models with Group-Specific Terms with
#' `rstanarm`" \url{http://mc-stan.org/rstanarm/articles/glmer.html}
#'
#' @examplesIf rlang::is_installed("modeldata")
#' library(recipes)
#' library(dplyr)
#' library(modeldata)
#'
#' data(grants)
#'
#' set.seed(1)
#' grants_other <- sample_n(grants_other, 500)
#' \donttest{
#' reencoded <- recipe(class ~ sponsor_code, data = grants_other) %>%
#'   step_lencode_bayes(sponsor_code, outcome = vars(class))
#' }
#' @export
step_lencode_bayes <-
  function(recipe,
           ...,
           role = NA,
           trained = FALSE,
           outcome = NULL,
           options = list(seed = sample.int(10^5, 1)),
           verbose = FALSE,
           mapping = NULL,
           skip = FALSE,
           id = rand_id("lencode_bayes")) {
    if (is.null(outcome)) {
      rlang::abort("Please list a variable in `outcome`")
    }
    add_step(
      recipe,
      step_lencode_bayes_new(
        terms = enquos(...),
        role = role,
        trained = trained,
        outcome = outcome,
        options = options,
        verbose = verbose,
        mapping = mapping,
        skip = skip,
        id = id,
        case_weights = NULL
      )
    )
  }

step_lencode_bayes_new <-
  function(terms, role, trained, outcome, options, verbose, mapping, skip, id,
           case_weights) {
    step(
      subclass = "lencode_bayes",
      terms = terms,
      role = role,
      trained = trained,
      outcome = outcome,
      options = options,
      verbose = verbose,
      mapping = mapping,
      skip = skip,
      id = id,
      case_weights = case_weights
    )
  }

#' @export
prep.step_lencode_bayes <- function(x, training, info = NULL, ...) {
  col_names <- recipes_eval_select(x$terms, training, info)

  wts <- get_case_weights(info, training)
  were_weights_used <- are_weights_used(wts)
  if (isFALSE(were_weights_used) || is.null(wts)) {
    wts <- NULL
  }

  if (length(col_names) > 0) {
    check_type(training[, col_names], types = c("string", "factor", "ordered"))
    y_name <- recipes_eval_select(x$outcome, training, info)
    res <-
      purrr::map(training[, col_names], stan_coefs,
        y = training[, y_name],
        x$options, x$verbose, wts
      )
  } else {
    res <- list()
  }

  step_lencode_bayes_new(
    terms = x$terms,
    role = x$role,
    trained = TRUE,
    outcome = x$outcome,
    options = x$options,
    verbose = x$verbose,
    mapping = res,
    skip = x$skip,
    id = x$id,
    case_weights = were_weights_used
  )
}

stan_coefs <- function(x, y, options, verbose, wts = NULL, ...) {
  rlang::check_installed("rstanarm")
  if (is.null(options$family)) {
    if (is.factor(y[[1]])) {
      fam <- binomial()
    } else {
      fam <- gaussian()
    }
  } else {
    fam <- options$family
    options$family <- NULL
  }

  form <- as.formula(paste0(names(y), "~ (1|value)"))

  if (is.vector(x) || is.factor(x)) {
    x <- tibble(value = x)
  } else {
    x <- as_tibble(x)
  }

  args <-
    list(
      form,
      data = vec_cbind(x, y),
      family = fam,
      na.action = na.omit
    )
  if (length(options) > 0) {
    args <- c(args, options)
  }
  if (!is.null(wts)) {
    args$weights <- as.double(wts)
  }

  cl <- rlang::call2("stan_glmer", .ns = "rstanarm", !!!args)

  if (!verbose) {
    junk <- capture.output(mod <- rlang::eval_tidy(cl))
  } else {
    mod <- rlang::eval_tidy(cl)
  }

  coefs <- coef(mod)$value
  coefs <- as.data.frame(coefs)
  coefs <- set_names(coefs, "..value")
  coefs <- rownames_to_column(coefs, "..level")
  coefs <- as_tibble(coefs)
  mean_coef <- mean(coefs$..value, na.rm = TRUE, trim = .1)
  coefs$..value[is.na(coefs$..value)] <- mean_coef
  new_row <- tibble(..level = "..new", ..value = mean_coef)
  coefs <- bind_rows(coefs, new_row)
  if (is.factor(y[[1]])) {
    coefs$..value <- -coefs$..value
  }
  coefs
}

#' @export
bake.step_lencode_bayes <- function(object, new_data, ...) {
  col_names <- names(object$mapping)
  check_new_data(col_names, object, new_data)

  for (col_name in col_names) {
    new_data[[col_name]] <- map_glm_coef(
      new_data[, col_name], # map_glm_coef() expects a tibble
      object$mapping[[col_name]]
    )
  }

  new_data
}

#' @export
print.step_lencode_bayes <-
  function(x, width = max(20, options()$width - 31), ...) {
    title <- "Linear embedding for factors via Bayesian GLM for "
    print_step(
      names(x$mapping), x$terms, x$trained, title, width,
      case_weights = x$case_weights
    )
    invisible(x)
  }

  #' @rdname step_lencode_bayes
  #' @usage NULL
  #' @export
tidy.step_lencode_bayes <- function(x, ...) {
  if (is_trained(x)) {
    if (length(x$mapping) == 0) {
      res <- tibble(
        terms = character(),
        level = character(),
        value = double()
      )
    } else {
      for (i in seq_along(x$mapping)) {
        x$mapping[[i]]$terms <- names(x$mapping)[i]
      }
      res <- bind_rows(x$mapping)
      names(res) <- gsub("^\\.\\.", "", names(res))
    }
  } else {
    term_names <- sel2char(x$terms)
    res <- tibble(
      terms = term_names,
      level = rep(na_chr, length(term_names)),
      value = rep(na_dbl, length(term_names))
    )
  }
  res$id <- x$id
  res
}

#' S3 methods for tracking which additional packages are needed for steps.
#'
#' Recipe-adjacent packages always list themselves as a required package so that
#' the steps can function properly within parallel processing schemes.
#' @param x A recipe step
#' @return A character vector
#' @rdname required_pkgs.embed
#' @keywords internal
#' @export
required_pkgs.step_lencode_bayes <- function(x, ...) {
  c("rstanarm", "embed")
}