R/s_GLMNET.R

In egenn/rtemis: Machine Learning and Visualization

# s_GLMNET.R
# ::rtemis::
# 2016-7 E.D. Gennatas www.lambdamd.org

#' GLM with Elastic Net Regularization \[C, R, S\]
#'
#' Train an elastic net model
#'
#' `s_GLMNET` runs `glmnet::cv.glmnet` for each value of alpha, for each resample in
#' `grid.resample.params`.
#' Mean values for `min.lambda` and MSE (Regression) or Accuracy (Classification) are aggregated for each
#' alpha and resample combination
#'
#' `\[gS\]` Indicates tunable hyperparameters: If more than a single value is provided, grid search will be
#' automatically performed
#'
#' @inheritParams s_GLM
#' @inheritParams s_CART
#' @param alpha \[gS\] Float \[0, 1\]: The elasticnet mixing parameter:
#'   `a = 0` is the ridge penalty, `a = 1` is the lasso penalty
#' @param lambda \[gS\] Float vector: Best left to NULL, `cv.glmnet` will
#' compute its own lambda sequence
#' @param nlambda Integer: Number of lambda values to compute
#' @param which.cv.lambda Character: Which lambda to use for prediction:
#'  "lambda.1se" or "lambda.min"
#' @param penalty.factor Float vector: Multiply the penalty for each coefficient by
#' the values in this vector. This is most useful for specifying different penalties
#' for different groups of variables
#' @param intercept Logical: If TRUE, include intercept in the model.
#' @param nway.interactions Integer: Number of n-way interactions to include in the model.
#' @param res.summary.fn Function: Used to average resample runs.
#' @param .gs (Internal use only)
#'
#' @author E.D. Gennatas
#' @seealso [train_cv] for external cross-validation
#' @family Supervised Learning
#' @family Interpretable models
#' @export
s_GLMNET <- function(x, y = NULL,
                     x.test = NULL, y.test = NULL,
                     x.name = NULL, y.name = NULL,
                     grid.resample.params = setup.resample("kfold", 5),
                     gridsearch.type = c("exhaustive", "randomized"),
                     gridsearch.randomized.p = .1,
                     intercept = TRUE,
                     nway.interactions = 0,
                     family = NULL,
                     alpha = seq(0, 1, .2),
                     lambda = NULL,
                     nlambda = 100,
                     which.cv.lambda = c("lambda.1se", "lambda.min"),
                     penalty.factor = NULL,
                     weights = NULL,
                     ifw = TRUE,
                     ifw.type = 2,
                     upsample = FALSE,
                     downsample = FALSE,
                     resample.seed = NULL,
                     res.summary.fn = mean,
                     metric = NULL,
                     maximize = NULL,
                     .gs = FALSE,
                     n.cores = rtCores,
                     print.plot = FALSE,
                     plot.fitted = NULL,
                     plot.predicted = NULL,
                     plot.theme = rtTheme,
                     question = NULL,
                     verbose = TRUE,
                     outdir = NULL,
                     save.mod = ifelse(!is.null(outdir), TRUE, FALSE), ...) {
  # Intro ----
  if (missing(x)) {
    print(args(s_GLMNET))
    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 <- "GLMNET"

  # Dependencies ----
  dependency_check("glmnet")

  # Arguments ----
  if (missing(x)) {
    print(args(s_GLMNET))
    stop("x is missing")
  }
  if (is.null(y) && NCOL(x) < 2) {
    print(args(s_GLMNET))
    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)
  if (!is.null(outdir)) {
    outdir <- paste0(normalizePath(outdir, mustWork = FALSE), "/")
  }
  which.cv.lambda <- match.arg(which.cv.lambda)
  gridsearch.type <- match.arg(gridsearch.type)

  # 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
  .weights <- if (is.null(weights) && ifw) dt$weights else weights
  if (is.null(.weights)) .weights <- rep(1, NROW(y))
  if (is.null(family)) {
    if (type == "Regression") {
      family <- "gaussian"
    } else if (type == "Classification") {
      family <- if (length(levels(y)) == 2) "binomial" else "multinomial"
    } else if (type == "Survival") {
      family <- "cox"
    }
  } else {
    if (family %in% c("binomial", "multinomial") && type != "Classification") {
      y <- factor(y)
      if (!is.null(y.test)) y.test <- factor(y.test)
      type <- "Classification"
    }
  }
  # Cox does not have an intercept (it is part of the baseline hazard)
  if (type == "Survival") intercept <- FALSE
  if (verbose) dataSummary(x, y, x.test, y.test, type)

  if (!is.null(family) && family %in% c("binomial", "multinomial") && !is.factor(y)) {
    if (type == "Survival") {
      colnames(y) <- c("time", "status")
      if (!is.null(y.test)) colnames(y.test) <- c("time", "status")
    }
  }
  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
  }

  # Model matrix ----
  dat <- data.frame(x, y = y)
  if (nway.interactions > 0) {
    .formula <- as.formula(paste0("y ~ .^", nway.interactions))
    x <- model.matrix(.formula, dat)[, -1]
  } else {
    .formula <- as.formula("y ~ .")
    x <- model.matrix(.formula, dat)[, -1]
  }

  if (is.null(penalty.factor)) penalty.factor <- rep(1, NCOL(x))

  if (!is.null(x.test)) {
    # for model.matrix to work, add y if not provided
    y.test1 <- if (is.null(y.test)) sample(y, NROW(x.test)) else y.test
    dat.test <- data.frame(x.test, y = y.test1)
    x.test <- model.matrix(.formula, dat.test)[, -1]
  }

  # Grid Search ----
  if (is.null(metric)) {
    if (type == "Classification") {
      metric <- "Balanced Accuracy"
    } else if (type == "Regression") {
      metric <- "MSE"
    }
  }
  if (is.null(maximize)) {
    maximize <- if (type == "Classification") TRUE else FALSE
  }
  cv.lambda <- is.null(lambda)
  do.gs <- is.null(lambda) | length(alpha) > 1 | length(lambda) > 1
  if (!.gs && do.gs) {
    gs <- gridSearchLearn(x, y,
      mod.name,
      resample.params = grid.resample.params,
      grid.params = list(
        alpha = alpha,
        lambda = lambda
      ),
      fixed.params = list(
        .gs = TRUE,
        which.cv.lambda = which.cv.lambda
      ),
      search.type = gridsearch.type,
      randomized.p = gridsearch.randomized.p,
      weights = weights,
      metric = metric,
      maximize = maximize,
      verbose = verbose,
      n.cores = n.cores
    )
    alpha <- gs$best.tune$alpha
    lambda <- gs$best.tune$lambda
  } else {
    gs <- NULL
  }
  if (verbose) {
    parameterSummary(alpha, lambda,
      newline.pre = TRUE
    )
  }

  # glmnet::cv.glmnet/glmnet ----
  if (.gs && cv.lambda) {
    mod <- glmnet::cv.glmnet(x,
      if (family == "binomial") reverseLevels(y) else y,
      family = family,
      alpha = alpha,
      lambda = lambda,
      nlambda = nlambda,
      weights = .weights,
      intercept = intercept,
      penalty.factor = penalty.factor, ...
    )
  } else {
    if (verbose) msg2("Training elastic net model...", newline.pre = TRUE)
    mod <- glmnet::glmnet(x,
      if (family == "binomial") reverseLevels(y) else y,
      family = family,
      alpha = alpha,
      lambda = lambda,
      nlambda = nlambda,
      weights = .weights,
      intercept = intercept,
      penalty.factor = penalty.factor, ...
    )
  }

  # Fitted ----
  if (type == "Regression" || type == "Survival") {
    fitted <- as.numeric(predict(mod, newx = x))
    fitted.prob <- NULL
  } else {
    if (family == "binomial") {
      fitted.prob <- predict(mod, x, type = "response")[, 1]
      fitted <- factor(ifelse(fitted.prob >= .5, 1, 0), levels = c(1, 0))
      levels(fitted) <- levels(y)
    } else {
      fitted.prob <- predict(mod, x, type = "response")
      fitted <- factor(colnames(fitted.prob)[apply(fitted.prob, 1, which.max)],
        levels = levels(y)
      )
    }
  }

  error.train <- mod_error(y, fitted, fitted.prob)
  if (verbose) errorSummary(error.train, mod.name)

  # Predicted ----
  predicted <- predicted.prob <- error.test <- NULL
  if (!is.null(x.test)) {
    if (type == "Regression" || type == "Survival") {
      predicted <- as.numeric(predict(mod, newx = x.test))
      predicted.prob <- NULL
    } else {
      if (family == "binomial") {
        predicted.prob <- predict(mod, x.test, type = "response")[, 1]
        predicted <- factor(ifelse(predicted.prob >= .5, 1, 0), levels = c(1, 0))
        levels(predicted) <- levels(y)
      } else {
        predicted.prob <- predict(mod, x.test, type = "response")
        predicted <- factor(colnames(predicted.prob)[apply(predicted.prob, 1, which.max)],
          levels = levels(y)
        )
      }
    }

    if (!is.null(y.test)) {
      error.test <- mod_error(y.test, predicted, predicted.prob)
      if (verbose) errorSummary(error.test, mod.name)
    }
  }

  # Outro ----
  rt <- rtModSet(
    rtclass = "rtMod",
    mod = mod,
    mod.name = mod.name,
    type = type,
    gridsearch = gs,
    parameters = list(lambda = lambda, alpha = alpha),
    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.bag = NULL,
    se.fit = NULL,
    error.train = error.train,
    predicted = predicted,
    predicted.prob = predicted.prob,
    se.prediction = NULL,
    error.test = error.test,
    varimp = as.matrix(coef(mod))[-1, 1],
    question = question
  )

  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_GLMNET