diff --git a/R-package/R/callbacks.R b/R-package/R/callbacks.R index b3d6bdb1ae0a..02e0a7cd4b8e 100644 --- a/R-package/R/callbacks.R +++ b/R-package/R/callbacks.R @@ -280,7 +280,6 @@ cb.reset.parameters <- function(new_params) { #' \code{iteration}, #' \code{begin_iteration}, #' \code{end_iteration}, -#' \code{num_parallel_tree}. #' #' @seealso #' \code{\link{callbacks}}, @@ -291,7 +290,6 @@ cb.early.stop <- function(stopping_rounds, maximize = FALSE, metric_name = NULL, verbose = TRUE) { # state variables best_iteration <- -1 - best_ntreelimit <- -1 best_score <- Inf best_msg <- NULL metric_idx <- 1 @@ -358,12 +356,10 @@ cb.early.stop <- function(stopping_rounds, maximize = FALSE, # If the difference is due to floating-point truncation, update best_score best_score <- attr_best_score } - xgb.attr(env$bst, "best_iteration") <- best_iteration - xgb.attr(env$bst, "best_ntreelimit") <- best_ntreelimit + xgb.attr(env$bst, "best_iteration") <- best_iteration - 1 xgb.attr(env$bst, "best_score") <- best_score } else { env$basket$best_iteration <- best_iteration - env$basket$best_ntreelimit <- best_ntreelimit } } @@ -385,14 +381,13 @@ cb.early.stop <- function(stopping_rounds, maximize = FALSE, ) best_score <<- score best_iteration <<- i - best_ntreelimit <<- best_iteration * env$num_parallel_tree # save the property to attributes, so they will occur in checkpoint if (!is.null(env$bst)) { xgb.attributes(env$bst) <- list( best_iteration = best_iteration - 1, # convert to 0-based index best_score = best_score, - best_msg = best_msg, - best_ntreelimit = best_ntreelimit) + best_msg = best_msg + ) } } else if (i - best_iteration >= stopping_rounds) { env$stop_condition <- TRUE @@ -475,8 +470,6 @@ cb.save.model <- function(save_period = 0, save_name = "xgboost.ubj") { #' \code{data}, #' \code{end_iteration}, #' \code{params}, -#' \code{num_parallel_tree}, -#' \code{num_class}. #' #' @return #' Predictions are returned inside of the \code{pred} element, which is either a vector or a matrix, @@ -499,19 +492,21 @@ cb.cv.predict <- function(save_models = FALSE) { stop("'cb.cv.predict' callback requires 'basket' and 'bst_folds' lists in its calling frame") N <- nrow(env$data) - pred <- - if (env$num_class > 1) { - matrix(NA_real_, N, env$num_class) - } else { - rep(NA_real_, N) - } + pred <- NULL - iterationrange <- c(1, NVL(env$basket$best_iteration, env$end_iteration) + 1) + iterationrange <- c(1, NVL(env$basket$best_iteration, env$end_iteration)) if (NVL(env$params[['booster']], '') == 'gblinear') { - iterationrange <- c(1, 1) # must be 0 for gblinear + iterationrange <- "all" } for (fd in env$bst_folds) { pr <- predict(fd$bst, fd$watchlist[[2]], iterationrange = iterationrange, reshape = TRUE) + if (is.null(pred)) { + if (NCOL(pr) > 1L) { + pred <- matrix(NA_real_, N, ncol(pr)) + } else { + pred <- matrix(NA_real_, N) + } + } if (is.matrix(pred)) { pred[fd$index, ] <- pr } else { diff --git a/R-package/R/utils.R b/R-package/R/utils.R index 945d86132a08..e8ae787fc722 100644 --- a/R-package/R/utils.R +++ b/R-package/R/utils.R @@ -208,7 +208,7 @@ xgb.iter.eval <- function(bst, watchlist, iter, feval) { res <- sapply(seq_along(watchlist), function(j) { w <- watchlist[[j]] ## predict using all trees - preds <- predict(bst, w, outputmargin = TRUE, iterationrange = c(1, 1)) + preds <- predict(bst, w, outputmargin = TRUE, iterationrange = "all") eval_res <- feval(preds, w) out <- eval_res$value names(out) <- paste0(evnames[j], "-", eval_res$metric) diff --git a/R-package/R/xgb.Booster.R b/R-package/R/xgb.Booster.R index cee7e9fc5887..2575c369031b 100644 --- a/R-package/R/xgb.Booster.R +++ b/R-package/R/xgb.Booster.R @@ -89,7 +89,6 @@ xgb.get.handle <- function(object) { #' @param outputmargin Whether the prediction should be returned in the form of original untransformed #' sum of predictions from boosting iterations' results. E.g., setting `outputmargin=TRUE` for #' logistic regression would return log-odds instead of probabilities. -#' @param ntreelimit Deprecated, use `iterationrange` instead. #' @param predleaf Whether to predict pre-tree leaf indices. #' @param predcontrib Whether to return feature contributions to individual predictions (see Details). #' @param approxcontrib Whether to use a fast approximation for feature contributions (see Details). @@ -99,11 +98,17 @@ xgb.get.handle <- function(object) { #' or `predinteraction` is `TRUE`. #' @param training Whether the predictions are used for training. For dart booster, #' training predicting will perform dropout. -#' @param iterationrange Specifies which trees are used in prediction. For -#' example, take a random forest with 100 rounds. -#' With `iterationrange=c(1, 21)`, only the trees built during `[1, 21)` (half open set) -#' rounds are used in this prediction. The index is 1-based just like an R vector. When set -#' to `c(1, 1)`, XGBoost will use all trees. +#' @param iterationrange Sequence of rounds/iterations from the model to use for prediction, specified by passing +#' a two-dimensional vector with the start and end numbers in the sequence (same format as R's `seq` - i.e. +#' base-1 indexing, and inclusive of both ends). +#' +#' For example, passing `c(1,20)` will predict using the first twenty iterations, while passing `c(1,1)` will +#' predict using only the first one. +#' +#' If passing `NULL`, will either stop at the best iteration if the model used early stopping, or use all +#' of the iterations (rounds) otherwise. +#' +#' If passing "all", will use all of the rounds regardless of whether the model had early stopping or not. #' @param strict_shape Default is `FALSE`. When set to `TRUE`, the output #' type and shape of predictions are invariant to the model type. #' @param ... Not used. @@ -189,7 +194,7 @@ xgb.get.handle <- function(object) { #' # use all trees by default #' pred <- predict(bst, test$data) #' # use only the 1st tree -#' pred1 <- predict(bst, test$data, iterationrange = c(1, 2)) +#' pred1 <- predict(bst, test$data, iterationrange = c(1, 1)) #' #' # Predicting tree leafs: #' # the result is an nsamples X ntrees matrix @@ -260,11 +265,11 @@ xgb.get.handle <- function(object) { #' all.equal(pred, pred_labels) #' # prediction from using only 5 iterations should result #' # in the same error as seen in iteration 5: -#' pred5 <- predict(bst, as.matrix(iris[, -5]), iterationrange = c(1, 6)) +#' pred5 <- predict(bst, as.matrix(iris[, -5]), iterationrange = c(1, 5)) #' sum(pred5 != lb) / length(lb) #' #' @export -predict.xgb.Booster <- function(object, newdata, missing = NA, outputmargin = FALSE, ntreelimit = NULL, +predict.xgb.Booster <- function(object, newdata, missing = NA, outputmargin = FALSE, predleaf = FALSE, predcontrib = FALSE, approxcontrib = FALSE, predinteraction = FALSE, reshape = FALSE, training = FALSE, iterationrange = NULL, strict_shape = FALSE, ...) { if (!inherits(newdata, "xgb.DMatrix")) { @@ -275,25 +280,22 @@ predict.xgb.Booster <- function(object, newdata, missing = NA, outputmargin = FA ) } - if (NVL(xgb.booster_type(object), '') == 'gblinear' || is.null(ntreelimit)) - ntreelimit <- 0 - if (ntreelimit != 0 && is.null(iterationrange)) { - ## only ntreelimit, initialize iteration range - iterationrange <- c(0, 0) - } else if (ntreelimit == 0 && !is.null(iterationrange)) { - ## only iteration range, handle 1-based indexing - iterationrange <- c(iterationrange[1] - 1, iterationrange[2] - 1) - } else if (ntreelimit != 0 && !is.null(iterationrange)) { - ## both are specified, let libgxgboost throw an error + if (!is.null(iterationrange)) { + if (is.character(iterationrange) && + length(iterationrange) == 1 && + iterationrange == "all") { + iterationrange <- c(0, 0) + } else { + iterationrange[1] <- iterationrange[1] - 1 # base-0 indexing + } } else { ## no limit is supplied, use best best_iteration <- xgb.best_iteration(object) if (is.null(best_iteration)) { iterationrange <- c(0, 0) } else { - ## We don't need to + 1 as R is 1-based index. - iterationrange <- c(0, as.integer(best_iteration)) + iterationrange <- c(0, as.integer(best_iteration) + 1L) } } ## Handle the 0 length values. @@ -312,7 +314,6 @@ predict.xgb.Booster <- function(object, newdata, missing = NA, outputmargin = FA strict_shape = box(TRUE), iteration_begin = box(as.integer(iterationrange[1])), iteration_end = box(as.integer(iterationrange[2])), - ntree_limit = box(as.integer(ntreelimit)), type = box(as.integer(0)) ) @@ -492,7 +493,7 @@ xgb.attr <- function(object, name) { return(NULL) } if (!is.null(out)) { - if (name %in% c("best_iteration", "best_ntreelimit", "best_score")) { + if (name %in% c("best_iteration", "best_score")) { out <- as.numeric(out) } } @@ -710,12 +711,6 @@ variable.names.xgb.Booster <- function(object, ...) { return(getinfo(object, "feature_name")) } -xgb.ntree <- function(bst) { - config <- xgb.config(bst) - out <- strtoi(config$learner$gradient_booster$gbtree_model_param$num_trees) - return(out) -} - xgb.nthread <- function(bst) { config <- xgb.config(bst) out <- strtoi(config$learner$generic_param$nthread) diff --git a/R-package/R/xgb.cv.R b/R-package/R/xgb.cv.R index a960957ca313..eb0495631d6e 100644 --- a/R-package/R/xgb.cv.R +++ b/R-package/R/xgb.cv.R @@ -103,7 +103,6 @@ #' parameter or randomly generated. #' \item \code{best_iteration} iteration number with the best evaluation metric value #' (only available with early stopping). -#' \item \code{best_ntreelimit} and the \code{ntreelimit} Deprecated attributes, use \code{best_iteration} instead. #' \item \code{pred} CV prediction values available when \code{prediction} is set. #' It is either vector or matrix (see \code{\link{cb.cv.predict}}). #' \item \code{models} a list of the CV folds' models. It is only available with the explicit @@ -218,7 +217,6 @@ xgb.cv <- function(params = list(), data, nrounds, nfold, label = NULL, missing # extract parameters that can affect the relationship b/w #trees and #iterations num_class <- max(as.numeric(NVL(params[['num_class']], 1)), 1) # nolint - num_parallel_tree <- max(as.numeric(NVL(params[['num_parallel_tree']], 1)), 1) # nolint # those are fixed for CV (no training continuation) begin_iteration <- 1 @@ -318,7 +316,7 @@ print.xgb.cv.synchronous <- function(x, verbose = FALSE, ...) { }) } - for (n in c('niter', 'best_iteration', 'best_ntreelimit')) { + for (n in c('niter', 'best_iteration')) { if (is.null(x[[n]])) next cat(n, ': ', x[[n]], '\n', sep = '') diff --git a/R-package/R/xgb.train.R b/R-package/R/xgb.train.R index a313ed32f414..f0f2332b58c3 100644 --- a/R-package/R/xgb.train.R +++ b/R-package/R/xgb.train.R @@ -393,7 +393,6 @@ xgb.train <- function(params = list(), data, nrounds, watchlist = list(), # Note: it might look like these aren't used, but they need to be defined in this # environment for the callbacks for work correctly. num_class <- max(as.numeric(NVL(params[['num_class']], 1)), 1) # nolint - num_parallel_tree <- max(as.numeric(NVL(params[['num_parallel_tree']], 1)), 1) # nolint if (is_update && nrounds > niter_init) stop("nrounds cannot be larger than ", niter_init, " (nrounds of xgb_model)") diff --git a/R-package/demo/predict_first_ntree.R b/R-package/demo/predict_first_ntree.R index 02c168b77e43..179c18c707f4 100644 --- a/R-package/demo/predict_first_ntree.R +++ b/R-package/demo/predict_first_ntree.R @@ -15,7 +15,7 @@ cat('start testing prediction from first n trees\n') labels <- getinfo(dtest, 'label') ### predict using first 1 tree -ypred1 <- predict(bst, dtest, ntreelimit = 1) +ypred1 <- predict(bst, dtest, iterationrange = c(1, 1)) # by default, we predict using all the trees ypred2 <- predict(bst, dtest) diff --git a/R-package/man/cb.cv.predict.Rd b/R-package/man/cb.cv.predict.Rd index ded899e8a4b1..4cabac1c9569 100644 --- a/R-package/man/cb.cv.predict.Rd +++ b/R-package/man/cb.cv.predict.Rd @@ -35,8 +35,6 @@ Callback function expects the following values to be set in its calling frame: \code{data}, \code{end_iteration}, \code{params}, -\code{num_parallel_tree}, -\code{num_class}. } \seealso{ \code{\link{callbacks}} diff --git a/R-package/man/cb.early.stop.Rd b/R-package/man/cb.early.stop.Rd index 7b6efa8427a2..7cd51a3ce563 100644 --- a/R-package/man/cb.early.stop.Rd +++ b/R-package/man/cb.early.stop.Rd @@ -55,7 +55,6 @@ Callback function expects the following values to be set in its calling frame: \code{iteration}, \code{begin_iteration}, \code{end_iteration}, -\code{num_parallel_tree}. } \seealso{ \code{\link{callbacks}}, diff --git a/R-package/man/predict.xgb.Booster.Rd b/R-package/man/predict.xgb.Booster.Rd index 66194c64fbec..7a6dd6c1306b 100644 --- a/R-package/man/predict.xgb.Booster.Rd +++ b/R-package/man/predict.xgb.Booster.Rd @@ -9,7 +9,6 @@ newdata, missing = NA, outputmargin = FALSE, - ntreelimit = NULL, predleaf = FALSE, predcontrib = FALSE, approxcontrib = FALSE, @@ -36,8 +35,6 @@ missing values in data (e.g., 0 or some other extreme value).} sum of predictions from boosting iterations' results. E.g., setting \code{outputmargin=TRUE} for logistic regression would return log-odds instead of probabilities.} -\item{ntreelimit}{Deprecated, use \code{iterationrange} instead.} - \item{predleaf}{Whether to predict pre-tree leaf indices.} \item{predcontrib}{Whether to return feature contributions to individual predictions (see Details).} @@ -53,11 +50,18 @@ or \code{predinteraction} is \code{TRUE}.} \item{training}{Whether the predictions are used for training. For dart booster, training predicting will perform dropout.} -\item{iterationrange}{Specifies which trees are used in prediction. For -example, take a random forest with 100 rounds. -With \code{iterationrange=c(1, 21)}, only the trees built during \verb{[1, 21)} (half open set) -rounds are used in this prediction. The index is 1-based just like an R vector. When set -to \code{c(1, 1)}, XGBoost will use all trees.} +\item{iterationrange}{Sequence of rounds/iterations from the model to use for prediction, specified by passing +a two-dimensional vector with the start and end numbers in the sequence (same format as R's \code{seq} - i.e. +base-1 indexing, and inclusive of both ends). + +\if{html}{\out{