R/s_BayesGLM.R

In egenn/rtemis: Machine Learning and Visualization

# s_BayesGLM.R
# ::rtemis::
# 2019 E.D. Gennatas www.lambdamd.org

#' Bayesian GLM
#'
#' Train a bayesian GLM using `arm::bayesglm`
#'
#' @inheritParams s_CART
#' @param family Character or function for the error distribution and link function to 
#' be used. See `arm::bayesglm` for details.
#' @param prior.mean Numeric, vector: Prior mean for the coefficients. If scalar,
#' it will be replicated to length N features.
#' @param prior.scale Numeric, vector: Prior scale for the coefficients. Default = NULL,
#' which results in 2.5 for logit, 2.5*1.6 for probit. If scalar,
#' it will be replicated to length N features.
#' @param prior.df Numeric: Prior degrees of freedom for the coefficients. Set to 1 for
#' t distribution; set to Inf for normal prior distribution. If scalar,
#' it will be replicated to length N features.
#' @param prior.mean.for.intercept Numeric: Prior mean for the intercept.
#' @param prior.scale.for.intercept Numeric: Default = NULL, which results in 10 for a 
#' logit model, and 10*1.6 for probit model.
#' @param prior.df.for.intercept Numeric: Prior df for the intercept.
#' @param min.prior.scale Numeric: Minimum prior scale for the coefficients.
#' @param scaled Logical: If TRUE, the scale for the prior distributions are:
#' For feature with single value, use `prior.scale`, for predictor with two values,
#' use `prior.scale/range(x)`, for more than two values, use `prior.scale/(2*sd(x))`.
#' If response is gaussian, `prior.scale` is multiplied by `2 * sd(y)`.
#' Default = TRUE
#' @param keep.order Logical: If TRUE, the feature positions are maintained, otherwise they are
#' reordered: main effects, interactions, second-order, third-order, etc.
#' @param drop.baseline Logical: If TRUE, drop the base level of factor features.
#' @param maxit Integer: Maximum number of iterations
#' @param ... Additional parameters to pass to `arm::bayesglm`
#' 
#' @family Bayesian
#' @family Supervised Learning
#' @export
#' @author E.D. Gennatas

s_BayesGLM <- function(x, y = NULL,
                       x.test = NULL, y.test = NULL,
                       family = NULL,
                       prior.mean = 0,
                       prior.scale = NULL,
                       prior.df = 1,
                       prior.mean.for.intercept = 0,
                       prior.scale.for.intercept = NULL,
                       prior.df.for.intercept = 1,
                       min.prior.scale = 1e-12,
                       scaled = TRUE,
                       keep.order = TRUE,
                       drop.baseline = TRUE,
                       maxit = 100,
                       x.name = NULL, y.name = NULL,
                       weights = NULL,
                       ifw = TRUE,
                       ifw.type = 2,
                       upsample = FALSE,
                       downsample = FALSE,
                       resample.seed = NULL,
                       metric = NULL,
                       maximize = NULL,
                       print.plot = FALSE,
                       plot.fitted = NULL,
                       plot.predicted = NULL,
                       plot.theme = rtTheme,
                       question = NULL,
                       grid.verbose = verbose,
                       verbose = TRUE,
                       outdir = NULL,
                       save.mod = ifelse(!is.null(outdir), TRUE, FALSE), ...) {

  # Intro ----
  if (missing(x)) {
    print(args(s_BayesGLM))
    return(invisible(9))
  }
  if (!is.null(outdir)) outdir <- paste0(normalizePath(outdir, mustWork = FALSE), "/")
  logFile <- if (!is.null(outdir)) {
    paste0(outdir, "/", sys.calls()[[1]][[1]], ".", format(Sys.time(), "%Y%m%d.%H%M%S"), ".log")
  } else {
    NULL
  }
  start.time <- intro(verbose = verbose, logFile = logFile)
  mod.name <- "BayesGLM"

  # Dependencies ----
  dependency_check("arm")

  # Arguments ----
  if (is.null(y) && NCOL(x) < 2) {
    print(args(s_BayesGLM))
    stop("y is missing")
  }
  if (is.null(x.name)) x.name <- getName(x, "x")
  if (is.null(y.name)) y.name <- getName(y, "y")
  if (!verbose) print.plot <- FALSE
  verbose <- verbose | !is.null(logFile)
  if (save.mod && is.null(outdir)) outdir <- paste0("./s.", mod.name)

  # Data ----
  dt <- prepare_data(x, y,
                    x.test, y.test,
                    ifw = ifw,
                    ifw.type = ifw.type,
                    upsample = upsample,
                    downsample = downsample,
                    resample.seed = resample.seed,
                    verbose = verbose)
  x <- dt$x
  y <- dt$y
  x.test <- dt$x.test
  y.test <- dt$y.test
  xnames <- dt$xnames
  type <- dt$type
  if (verbose) dataSummary(x, y, x.test, y.test, type)
  if (print.plot) {
    if (is.null(plot.fitted)) plot.fitted <- if (is.null(y.test)) TRUE else FALSE
    if (is.null(plot.predicted)) plot.predicted <- if (!is.null(y.test)) TRUE else FALSE
  } else {
    plot.fitted <- plot.predicted <- FALSE
  }
  if (type == "Classification") nlevels <- length(levels(y))

  if (is.null(metric)) {
    if (type == "Classification") {
      metric <- "Balanced Accuracy"
      if (is.null(maximize)) maximize <- TRUE
    } else if (type == "Regression") {
      metric <- "MSE"
      if (is.null(maximize)) maximize <- FALSE
    }
  }

  if (is.null(maximize)) {
    maximize <- if (type == "Classification") TRUE else FALSE
  }

  if (is.null(family)) {
    family <- if (type == "Regression") gaussian else binomial
  }

  # Formula ----
  df.train <- data.frame(x, y)
  colnames(df.train)[ncol(df.train)] <- y.name
  .formula <- as.formula(paste(y.name, "~ ."))

  extra.args <- list(...)
  parameters <- c(list(ifw = ifw,
                       ifw.type = ifw.type,
                       upsample = upsample,
                       resample.seed = resample.seed,
                       prior.mean = prior.mean,
                       prior.scale = prior.scale,
                       prior.df = prior.df,
                       prior.mean.for.intercept = prior.mean.for.intercept,
                       prior.scale.for.intercept = prior.scale.for.intercept,
                       prior.df.for.intercept = prior.df.for.intercept,
                       min.prior.scale = min.prior.scale,
                       scaled = scaled,
                       keep.order = keep.order,
                       drop.baseline = drop.baseline,
                       maxit = maxit), extra.args)

  # BayesGLM ----
  if (verbose) msg2("Training Bayesian GLM...", newline.pre = TRUE)
  args <- c(list(formula = .formula,
                 data = df.train,
                 family = family,
                 prior.mean = prior.mean,
                 prior.scale = prior.scale,
                 prior.df = prior.df,
                 prior.mean.for.intercept = prior.mean.for.intercept,
                 prior.scale.for.intercept = prior.scale.for.intercept,
                 prior.df.for.intercept = prior.df.for.intercept,
                 min.prior.scale = min.prior.scale,
                 scaled = scaled,
                 keep.order = keep.order,
                 drop.baseline = drop.baseline,
                 maxit = maxit),
            extra.args)
  mod <- do.call(arm::bayesglm, args)

  # Fitted ----
  if (type == "Classification") {
    fitted.prob <- 1 - predict(mod, type = "response")
    fitted <- factor(ifelse(fitted.prob >= .5, 1, 0), levels = c(1, 0))
    levels(fitted) <- levels(y)
  } else {
    fitted.prob <- NULL
    fitted <- predict(mod, x)
  }
  error.train <- mod_error(y, fitted, fitted.prob)
  if (verbose) errorSummary(error.train, mod.name)

  # Predicted ----
  predicted.prob <- NULL
  if (!is.null(x.test)) {
    if (type == "Classification") {
      predicted.prob <- 1 - predict(mod, x.test, type = "response")
      predicted <- factor(ifelse(predicted.prob >= .5, 1, 0), levels = c(1, 0))
      levels(predicted) <- levels(y)
    } else {
      predicted <- predict(mod, x.test)
    }
    if (!is.null(y.test)) {
      error.test <- mod_error(y.test, predicted, predicted.prob)
      if (verbose) errorSummary(error.test, mod.name)
    } else {
      error.test <- NULL
    }
  } else {
    predicted <- error.test <- NULL
  }

  # Outro ----
  extra <- list()
  rt <- rtModSet(rtclass = "rtMod",
                 mod = mod,
                 mod.name = mod.name,
                 type = type,
                 parameters = parameters,
                 y.train = y,
                 y.test = y.test,
                 x.name = x.name,
                 y.name = y.name,
                 xnames = xnames,
                 fitted = fitted,
                 fitted.prob = fitted.prob,
                 se.fit = NULL,
                 error.train = error.train,
                 predicted = predicted,
                 predicted.prob = predicted.prob,
                 se.prediction = NULL,
                 error.test = error.test,
                 varimp = mod$coefficients[-1],
                 question = question,
                 extra = extra)

  rtMod.out(rt,
            print.plot,
            plot.fitted,
            plot.predicted,
            y.test,
            mod.name,
            outdir,
            save.mod,
            verbose,
            plot.theme)

  outro(start.time, verbose = verbose, sinkOff = ifelse(is.null(logFile), FALSE, TRUE))
  rt

} # rtemis::s_BayesGLM