Fix [R-package] Prevent remembering parameters (#799)

* Revert "[R-package] Prevent remembering parameters (#796)"

This reverts commit c795e2c8e12e8fe989bde0ab4ee59c2301325814.

* Use model load/unload trick for free-ing up memory

* Missing comma

* Hand copy manual update (1/2)

* Hand copy manual update (2/2)

R-package/R/lgb.Dataset.R

 Dataset <- R6Class(
   classname = "lgb.Dataset",
-  cloneable = TRUE,
+  cloneable = FALSE,
   public = list(
     
-    # Logical to check whether a dataset can be used re-modeled in-memory as another Dataset or not
-    remodel = TRUE,
-    
     # Finalize will free up the handles
     finalize = function() {
       
@@ -279,9 +276,6 @@ Dataset <- R6Class(
         stop("lgb.Dataset.construct: label should be set")
       }
       
-      # Forcefully block construction
-      self$remodel <- FALSE
-      
       # Return self
       return(invisible(self))
       

R-package/R/lgb.train.R

@@ -27,6 +27,7 @@
 #'        If there's more than one, will check all of them
 #'        Returns the model with (best_iter + early_stopping_rounds)
 #'        If early stopping occurs, the model will have 'best_iter' field
+#' @param reset_data Boolean, setting it to TRUE (not the default value) will transform the booster model into a predictor model which frees up memory and the original datasets
 #' @param callbacks list of callback functions
 #'        List of callback functions that are applied at each iteration.
 #' @param ... other parameters, see parameters.md for more informations
@@ -70,6 +71,7 @@ lgb.train <- function(params = list(),
                       categorical_feature = NULL,
                       early_stopping_rounds = NULL,
                       callbacks = list(),
+                      reset_data = FALSE,
                       ...) {
   
   # Setup temporary variables
@@ -153,10 +155,7 @@ lgb.train <- function(params = list(),
   }
   
   # Construct datasets, if needed
-  if (data$remodel == TRUE) {
-    data <- data$clone(deep = FALSE)
-    data$construct()
-  }
+  data$construct()
   vaild_contain_train <- FALSE
   train_data_name <- "train"
   reduced_valid_sets <- list()
@@ -218,7 +217,7 @@ lgb.train <- function(params = list(),
   env$model <- booster
   env$begin_iteration <- begin_iteration
   env$end_iteration <- end_iteration
-
+  
   # Start training model using number of iterations to start and end with
   for (i in seq(from = begin_iteration, to = end_iteration)) {
     
@@ -262,6 +261,22 @@ lgb.train <- function(params = list(),
     
   }
   
+  # Check for booster model conversion to predictor model
+  if (reset_data) {
+    
+    # Store temporarily model data elsewhere
+    booster_old <- list(best_iter = booster$best_iter,
+                        best_score = booster$best_score,
+                        record_evals = booster$record_evals)
+    
+    # Reload model
+    booster <- lgb.load(model_str = booster$save_model_to_string())
+    booster$best_iter <- booster_old$best_iter
+    booster$best_score <- booster_old$best_score
+    booster$record_evals <- booster_old$record_evals
+    
+  }
+  
   # Return booster
   return(booster)
   

R-package/man/lgb.train.Rd

@@ -15,7 +15,8 @@ lgb.cv(params = list(), data, nrounds = 10, nfold = 3, label = NULL,
 lgb.train(params = list(), data, nrounds = 10, valids = list(),
   obj = NULL, eval = NULL, verbose = 1, record = TRUE, eval_freq = 1L,
   init_model = NULL, colnames = NULL, categorical_feature = NULL,
-  early_stopping_rounds = NULL, callbacks = list(), ...)
+  early_stopping_rounds = NULL, callbacks = list(), reset_data = FALSE,
+  ...)
 
 lightgbm(data, label = NULL, weight = NULL, params = list(),
   nrounds = 10, verbose = 1, eval_freq = 1L,
@@ -78,6 +79,8 @@ List of callback functions that are applied at each iteration.}
 
 \item{valids}{a list of \code{lgb.Dataset} objects, used for validation}
 
+\item{reset_data}{Boolean, setting it to TRUE (not the default value) will transform the booster model into a predictor model which frees up memory and the original datasets}
+
 \item{boosting}{boosting type. \code{gbdt}, \code{dart}}
 
 \item{num_leaves}{number of leaves in one tree. defaults to 127}

R-package/man/predict.lgb.Booster.Rd

@@ -5,7 +5,8 @@
 \title{Predict method for LightGBM model}
 \usage{
 \method{predict}{lgb.Booster}(object, data, num_iteration = NULL,
-  rawscore = FALSE, predleaf = FALSE, header = FALSE, reshape = FALSE)
+  rawscore = FALSE, predleaf = FALSE, header = FALSE, reshape = FALSE,
+  ...)
 }
 \arguments{
 \item{object}{Object of class \code{lgb.Booster}}