[R-package] consolidate testing constants in helpers file (#5992)

R-package/tests/testthat/helper.R

 # ref for this file:
 #
+# * https://r-pkgs.org/testing-design.html#testthat-helper-files
 # * https://r-pkgs.org/testing-design.html#testthat-setup-files
 
 # LightGBM-internal fix to comply with CRAN policy of only using up to 2 threads in tests and example.
@@ -10,3 +11,21 @@
 #   the check farm is a shared resource and will typically be running many checks simultaneously.
 #
 .LGB_MAX_THREADS <- 2L
+
+# by default, how much should results in tests be allowed to differ from hard-coded expected numbers?
+.LGB_NUMERIC_TOLERANCE <- 1e-6
+
+# are the tests running on Windows?
+.LGB_ON_WINDOWS <- .Platform$OS.type == "windows"
+.LGB_ON_32_BIT_WINDOWS <- .LGB_ON_WINDOWS && .Machine$sizeof.pointer != 8L
+
+# are the tests running in a UTF-8 locale?
+.LGB_UTF8_LOCALE <- all(endsWith(
+  Sys.getlocale(category = "LC_CTYPE")
+  , "UTF-8"
+))
+
+# control how many loud LightGBM's logger is in tests
+.LGB_VERBOSITY <- as.integer(
+  Sys.getenv("LIGHTGBM_TEST_VERBOSITY", "-1")
+)

R-package/tests/testthat/test_Predictor.R

 library(Matrix)
 
-VERBOSITY <- as.integer(
-  Sys.getenv("LIGHTGBM_TEST_VERBOSITY", "-1")
-)
-
-TOLERANCE <- 1e-6
-
 test_that("Predictor$finalize() should not fail", {
     X <- as.matrix(as.integer(iris[, "Species"]), ncol = 1L)
     y <- iris[["Sepal.Length"]]
@@ -16,7 +10,7 @@ test_that("Predictor$finalize() should not fail", {
             objective = "regression"
             , num_threads = .LGB_MAX_THREADS
         )
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , nrounds = 3L
     )
     model_file <- tempfile(fileext = ".model")
@@ -45,7 +39,7 @@ test_that("predictions do not fail for integer input", {
             objective = "regression"
             , num_threads = .LGB_MAX_THREADS
         )
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , nrounds = 3L
     )
     X_double <- X[c(1L, 51L, 101L), , drop = FALSE]
@@ -78,7 +72,7 @@ test_that("start_iteration works correctly", {
             num_leaves = 4L
             , learning_rate = 0.6
             , objective = "binary"
-            , verbosity = VERBOSITY
+            , verbosity = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = 50L
@@ -128,7 +122,7 @@ test_that("Feature contributions from sparse inputs produce sparse outputs", {
       data = dtrain
       , obj = "regression"
       , nrounds = 5L
-      , verbose = VERBOSITY
+      , verbose = .LGB_VERBOSITY
       , params = list(min_data_in_leaf = 5L, num_threads = .LGB_MAX_THREADS)
     )
 
@@ -159,7 +153,7 @@ test_that("Sparse feature contribution predictions do not take inputs with wrong
       data = dtrain
       , obj = "regression"
       , nrounds = 5L
-      , verbose = VERBOSITY
+      , verbose = .LGB_VERBOSITY
       , params = list(min_data_in_leaf = 5L, num_threads = .LGB_MAX_THREADS)
     )
 
@@ -189,7 +183,7 @@ test_that("Feature contribution predictions do not take non-general CSR or CSC i
       data = dtrain
       , obj = "regression"
       , nrounds = 5L
-      , verbose = VERBOSITY
+      , verbose = .LGB_VERBOSITY
       , params = list(min_data_in_leaf = 5L, num_threads = .LGB_MAX_THREADS)
     )
 
@@ -217,14 +211,14 @@ test_that("predict() params should override keyword argument for raw-score predi
       , num_threads = .LGB_MAX_THREADS
     )
     , nrounds = 10L
-    , verbose = VERBOSITY
+    , verbose = .LGB_VERBOSITY
   )
 
   # check that the predictions from predict.lgb.Booster() really look like raw score predictions
   preds_prob <- predict(bst, X)
   preds_raw_s3_keyword <- predict(bst, X, type = "raw")
   preds_prob_from_raw <- 1.0 / (1.0 + exp(-preds_raw_s3_keyword))
-  expect_equal(preds_prob, preds_prob_from_raw, tolerance = TOLERANCE)
+  expect_equal(preds_prob, preds_prob_from_raw, tolerance = .LGB_NUMERIC_TOLERANCE)
   accuracy <- sum(as.integer(preds_prob_from_raw > 0.5) == y) / length(y)
   expect_equal(accuracy, 1.0)
 
@@ -269,7 +263,7 @@ test_that("predict() params should override keyword argument for leaf-index pred
       , num_threads = .LGB_MAX_THREADS
     )
     , nrounds = 10L
-    , verbose = VERBOSITY
+    , verbose = .LGB_VERBOSITY
   )
 
   # check that predictions really look like leaf index predictions
@@ -323,7 +317,7 @@ test_that("predict() params should override keyword argument for feature contrib
       , num_threads = .LGB_MAX_THREADS
     )
     , nrounds = 10L
-    , verbose = VERBOSITY
+    , verbose = .LGB_VERBOSITY
   )
 
   # check that predictions really look like feature contributions
@@ -431,7 +425,7 @@ test_that("predict() keeps row names from data (regression)", {
         data = dtrain
         , obj = "regression"
         , nrounds = 5L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , params = list(min_data_in_leaf = 1L, num_threads = .LGB_MAX_THREADS)
     )
     .check_all_row_name_expectations(bst, X)
@@ -447,7 +441,7 @@ test_that("predict() keeps row names from data (binary classification)", {
         data = dtrain
         , obj = "binary"
         , nrounds = 5L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , params = list(num_threads = .LGB_MAX_THREADS)
     )
     .check_all_row_name_expectations(bst, X)
@@ -464,7 +458,7 @@ test_that("predict() keeps row names from data (multi-class classification)", {
         , obj = "multiclass"
         , params = list(num_class = 3L, num_threads = .LGB_MAX_THREADS)
         , nrounds = 5L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
     )
     .check_all_row_name_expectations(bst, X)
 })
@@ -485,7 +479,7 @@ test_that("predictions for regression and binary classification are returned as
       data = dtrain
       , obj = "regression"
       , nrounds = 5L
-      , verbose = VERBOSITY
+      , verbose = .LGB_VERBOSITY
       , params = list(min_data_in_leaf = 1L, num_threads = .LGB_MAX_THREADS)
     )
     pred <- predict(model, X)
@@ -503,7 +497,7 @@ test_that("predictions for regression and binary classification are returned as
       data = dtrain
       , obj = "binary"
       , nrounds = 5L
-      , verbose = VERBOSITY
+      , verbose = .LGB_VERBOSITY
       , params = list(num_threads = .LGB_MAX_THREADS)
     )
     pred <- predict(model, X)
@@ -523,7 +517,7 @@ test_that("predictions for multiclass classification are returned as matrix", {
       data = dtrain
       , obj = "multiclass"
       , nrounds = 5L
-      , verbose = VERBOSITY
+      , verbose = .LGB_VERBOSITY
       , params = list(num_class = 3L, num_threads = .LGB_MAX_THREADS)
     )
     pred <- predict(model, X)
@@ -668,7 +662,7 @@ test_that("predict type='class' returns predicted class for classification objec
         data = dtrain
         , obj = "binary"
         , nrounds = 5L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , params = list(num_threads = .LGB_MAX_THREADS)
     )
     pred <- predict(bst, X, type = "class")
@@ -682,7 +676,7 @@ test_that("predict type='class' returns predicted class for classification objec
       data = dtrain
       , obj = "multiclass"
       , nrounds = 5L
-      , verbose = VERBOSITY
+      , verbose = .LGB_VERBOSITY
       , params = list(num_class = 3L, num_threads = .LGB_MAX_THREADS)
     )
     pred <- predict(model, X, type = "class")
@@ -698,7 +692,7 @@ test_that("predict type='class' returns values in the target's range for regress
         data = dtrain
         , obj = "regression"
         , nrounds = 5L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , params = list(num_threads = .LGB_MAX_THREADS)
     )
     pred <- predict(bst, X, type = "class")

R-package/tests/testthat/test_custom_objective.R

-VERBOSITY <- as.integer(
-  Sys.getenv("LIGHTGBM_TEST_VERBOSITY", "-1")
-)
-
 data(agaricus.train, package = "lightgbm")
 data(agaricus.test, package = "lightgbm")
 dtrain <- lgb.Dataset(agaricus.train$data, label = agaricus.train$label)
 dtest <- lgb.Dataset(agaricus.test$data, label = agaricus.test$label)
 watchlist <- list(eval = dtest, train = dtrain)
 
-TOLERANCE <- 1e-6
-
 logregobj <- function(preds, dtrain) {
   labels <- get_field(dtrain, "label")
   preds <- 1.0 / (1.0 + exp(-preds))
@@ -38,7 +32,7 @@ param <- list(
   , learning_rate = 1.0
   , objective = logregobj
   , metric = "auc"
-  , verbose = VERBOSITY
+  , verbose = .LGB_VERBOSITY
   , num_threads = .LGB_MAX_THREADS
 )
 num_round <- 10L
@@ -54,7 +48,7 @@ test_that("using a custom objective, custom eval, and no other metrics works", {
     params = list(
       num_leaves = 8L
       , learning_rate = 1.0
-      , verbose = VERBOSITY
+      , verbose = .LGB_VERBOSITY
       , num_threads = .LGB_MAX_THREADS
     )
     , data = dtrain
@@ -65,11 +59,11 @@ test_that("using a custom objective, custom eval, and no other metrics works", {
   )
   expect_false(is.null(bst$record_evals))
   expect_equal(bst$best_iter, 4L)
-  expect_true(abs(bst$best_score - 0.000621) < TOLERANCE)
+  expect_true(abs(bst$best_score - 0.000621) < .LGB_NUMERIC_TOLERANCE)
 
   eval_results <- bst$eval_valid(feval = evalerror)[[1L]]
   expect_true(eval_results[["data_name"]] == "eval")
-  expect_true(abs(eval_results[["value"]] - 0.0006207325) < TOLERANCE)
+  expect_true(abs(eval_results[["value"]] - 0.0006207325) < .LGB_NUMERIC_TOLERANCE)
   expect_true(eval_results[["name"]] == "error")
   expect_false(eval_results[["higher_better"]])
 })
@@ -81,7 +75,7 @@ test_that("using a custom objective that returns wrong shape grad or hess raises
   bad_hess <- function(preds, dtrain) {
     return(list(grad = rep(1.0, length(preds)), hess = numeric(0L)))
   }
-  params <- list(num_leaves = 3L, verbose = VERBOSITY)
+  params <- list(num_leaves = 3L, verbose = .LGB_VERBOSITY)
   expect_error({
     lgb.train(params = params, data = dtrain, obj = bad_grad)
   }, sprintf("Expected custom objective function to return grad with length %d, got 0.", nrow(dtrain)))

R-package/tests/testthat/test_dataset.R

-VERBOSITY <- as.integer(
-  Sys.getenv("LIGHTGBM_TEST_VERBOSITY", "-1")
-)
-
 data(agaricus.train, package = "lightgbm")
 train_data <- agaricus.train$data[seq_len(1000L), ]
 train_label <- agaricus.train$label[seq_len(1000L)]
@@ -16,7 +12,7 @@ test_that("lgb.Dataset: basic construction, saving, loading", {
     test_data
     , label = test_label
     , params = list(
-      verbose = VERBOSITY
+      verbose = .LGB_VERBOSITY
     )
   )
   # from dense matrix
@@ -30,7 +26,7 @@ test_that("lgb.Dataset: basic construction, saving, loading", {
   dtest3 <- lgb.Dataset(
     tmp_file
     , params = list(
-      verbose = VERBOSITY
+      verbose = .LGB_VERBOSITY
     )
   )
   lgb.Dataset.construct(dtest3)
@@ -376,7 +372,7 @@ test_that("lgb.Dataset: should be able to run lgb.train() immediately after usin
     data = test_data
     , label = test_label
     , params = list(
-      verbose = VERBOSITY
+      verbose = .LGB_VERBOSITY
     )
   )
   tmp_file <- tempfile(pattern = "lgb.Dataset_")
@@ -393,7 +389,7 @@ test_that("lgb.Dataset: should be able to run lgb.train() immediately after usin
     , metric = "binary_logloss"
     , num_leaves = 5L
     , learning_rate = 1.0
-    , verbose = VERBOSITY
+    , verbose = .LGB_VERBOSITY
     , num_threads = .LGB_MAX_THREADS
   )
 
@@ -411,7 +407,7 @@ test_that("lgb.Dataset: should be able to run lgb.cv() immediately after using l
     data = test_data
     , label = test_label
     , params = list(
-      verbosity = VERBOSITY
+      verbosity = .LGB_VERBOSITY
     )
   )
   tmp_file <- tempfile(pattern = "lgb.Dataset_")
@@ -429,7 +425,7 @@ test_that("lgb.Dataset: should be able to run lgb.cv() immediately after using l
     , num_leaves = 5L
     , learning_rate = 1.0
     , num_iterations = 5L
-    , verbosity = VERBOSITY
+    , verbosity = .LGB_VERBOSITY
     , num_threads = .LGB_MAX_THREADS
   )
 
@@ -475,7 +471,7 @@ test_that("lgb.Dataset: should be able to create a Dataset from a text file with
     data = train_file
     , params = list(
       header = TRUE
-      , verbosity = VERBOSITY
+      , verbosity = .LGB_VERBOSITY
     )
   )
   dtrain$construct()
@@ -499,7 +495,7 @@ test_that("lgb.Dataset: should be able to create a Dataset from a text file with
     data = train_file
     , params = list(
       header = FALSE
-      , verbosity = VERBOSITY
+      , verbosity = .LGB_VERBOSITY
     )
   )
   dtrain$construct()

R-package/tests/testthat/test_learning_to_rank.R

-VERBOSITY <- as.integer(
-  Sys.getenv("LIGHTGBM_TEST_VERBOSITY", "-1")
-)
-
-# numerical tolerance to use when checking metric values
-TOLERANCE <- 1e-06
-
-ON_32_BIT_WINDOWS <- .Platform$OS.type == "windows" && .Machine$sizeof.pointer != 8L
-
 test_that("learning-to-rank with lgb.train() works as expected", {
     set.seed(708L)
     data(agaricus.train, package = "lightgbm")
@@ -26,7 +17,7 @@ test_that("learning-to-rank with lgb.train() works as expected", {
         , ndcg_at = ndcg_at
         , lambdarank_truncation_level = 3L
         , learning_rate = 0.001
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     model <- lgb.train(
@@ -60,15 +51,15 @@ test_that("learning-to-rank with lgb.train() works as expected", {
         , eval_names
     )
     expect_equal(eval_results[[1L]][["value"]], 0.775)
-    if (!ON_32_BIT_WINDOWS) {
-        expect_true(abs(eval_results[[2L]][["value"]] - 0.745986) < TOLERANCE)
-        expect_true(abs(eval_results[[3L]][["value"]] - 0.7351959) < TOLERANCE)
+    if (!.LGB_ON_32_BIT_WINDOWS) {
+        expect_true(abs(eval_results[[2L]][["value"]] - 0.745986) < .LGB_NUMERIC_TOLERANCE)
+        expect_true(abs(eval_results[[3L]][["value"]] - 0.7351959) < .LGB_NUMERIC_TOLERANCE)
     }
 })
 
 test_that("learning-to-rank with lgb.cv() works as expected", {
     testthat::skip_if(
-        ON_32_BIT_WINDOWS
+        .LGB_ON_32_BIT_WINDOWS
         , message = "Skipping on 32-bit Windows"
     )
     set.seed(708L)
@@ -91,7 +82,7 @@ test_that("learning-to-rank with lgb.cv() works as expected", {
         , label_gain = "0,1,3"
         , min_data = 1L
         , learning_rate = 0.01
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     nfold <- 4L
@@ -115,7 +106,7 @@ test_that("learning-to-rank with lgb.cv() works as expected", {
     best_score <- cv_bst$best_score
     expect_true(best_iter > 0L && best_iter <= nrounds)
     expect_true(best_score > 0.0 && best_score < 1.0)
-    expect_true(abs(best_score - 0.75) < TOLERANCE)
+    expect_true(abs(best_score - 0.75) < .LGB_NUMERIC_TOLERANCE)
 
     # best_score should be set for the first metric
     first_metric <- eval_names[[1L]]
@@ -138,19 +129,19 @@ test_that("learning-to-rank with lgb.cv() works as expected", {
 
     # first and last value of each metric should be as expected
     ndcg1_values <- c(0.675, 0.725, 0.65, 0.725, 0.75, 0.725, 0.75, 0.725, 0.75, 0.75)
-    expect_true(all(abs(unlist(eval_results[["ndcg@1"]][["eval"]]) - ndcg1_values) < TOLERANCE))
+    expect_true(all(abs(unlist(eval_results[["ndcg@1"]][["eval"]]) - ndcg1_values) < .LGB_NUMERIC_TOLERANCE))
 
     ndcg2_values <- c(
         0.6556574, 0.6669721, 0.6306574, 0.6476294, 0.6629581,
         0.6476294, 0.6629581, 0.6379581, 0.7113147, 0.6823008
     )
-    expect_true(all(abs(unlist(eval_results[["ndcg@2"]][["eval"]]) - ndcg2_values) < TOLERANCE))
+    expect_true(all(abs(unlist(eval_results[["ndcg@2"]][["eval"]]) - ndcg2_values) < .LGB_NUMERIC_TOLERANCE))
 
     ndcg3_values <- c(
         0.6484639, 0.6571238, 0.6469279, 0.6540516, 0.6481857,
         0.6481857, 0.6481857, 0.6466496, 0.7027939, 0.6629898
     )
-    expect_true(all(abs(unlist(eval_results[["ndcg@3"]][["eval"]]) - ndcg3_values) < TOLERANCE))
+    expect_true(all(abs(unlist(eval_results[["ndcg@3"]][["eval"]]) - ndcg3_values) < .LGB_NUMERIC_TOLERANCE))
 
     # check details of each booster
     for (bst in cv_bst$boosters) {

R-package/tests/testthat/test_lgb.Booster.R

-VERBOSITY <- as.integer(
-  Sys.getenv("LIGHTGBM_TEST_VERBOSITY", "-1")
-)
-
-ON_WINDOWS <- .Platform$OS.type == "windows"
-TOLERANCE <- 1e-6
-
 test_that("Booster$finalize() should not fail", {
     X <- as.matrix(as.integer(iris[, "Species"]), ncol = 1L)
     y <- iris[["Sepal.Length"]]
@@ -15,7 +8,7 @@ test_that("Booster$finalize() should not fail", {
             objective = "regression"
             , num_threads = .LGB_MAX_THREADS
         )
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , nrounds = 3L
     )
     expect_true(lgb.is.Booster(bst))
@@ -66,7 +59,7 @@ test_that("lgb.get.eval.result() should throw an informative error for incorrect
             , metric = "l2"
             , min_data = 1L
             , learning_rate = 1.0
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , data = dtrain
@@ -101,7 +94,7 @@ test_that("lgb.get.eval.result() should throw an informative error for incorrect
             , metric = "l2"
             , min_data = 1L
             , learning_rate = 1.0
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
         )
         , data = dtrain
         , nrounds = 5L
@@ -135,7 +128,7 @@ test_that("lgb.load() gives the expected error messages given different incorrec
             objective = "binary"
             , num_leaves = 4L
             , learning_rate = 1.0
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
         )
         , nrounds = 2L
     )
@@ -186,7 +179,7 @@ test_that("Loading a Booster from a text file works", {
         , metric = c("mape", "average_precision")
         , learning_rate = 1.0
         , objective = "binary"
-        , verbosity = VERBOSITY
+        , verbosity = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     bst <- lightgbm(
@@ -237,7 +230,7 @@ test_that("boosters with linear models at leaves can be written to text file and
         data = dtrain
         , nrounds = 10L
         , params = params
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
     )
     expect_true(lgb.is.Booster(bst))
 
@@ -271,7 +264,7 @@ test_that("Loading a Booster from a string works", {
             num_leaves = 4L
             , learning_rate = 1.0
             , objective = "binary"
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = 2L
@@ -307,7 +300,7 @@ test_that("Saving a large model to string should work", {
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = 500L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
     )
 
     pred <- predict(bst, train$data)
@@ -351,7 +344,7 @@ test_that("Saving a large model to JSON should work", {
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = 200L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
     )
 
     model_json <- bst$dump_model()
@@ -378,7 +371,7 @@ test_that("If a string and a file are both passed to lgb.load() the file is used
             num_leaves = 4L
             , learning_rate = 1.0
             , objective = "binary"
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = 2L
@@ -413,7 +406,7 @@ test_that("Creating a Booster from a Dataset should work", {
     bst <- Booster$new(
         params = list(
             objective = "binary"
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         ),
         train_set = dtrain
@@ -435,7 +428,7 @@ test_that("Creating a Booster from a Dataset with an existing predictor should w
             num_leaves = 4L
             , learning_rate = 1.0
             , objective = "binary"
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = nrounds
@@ -449,7 +442,7 @@ test_that("Creating a Booster from a Dataset with an existing predictor should w
     bst_from_ds <- Booster$new(
         train_set = dtest
         , params = list(
-            verbose = VERBOSITY
+            verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
     )
@@ -473,7 +466,7 @@ test_that("Booster$eval() should work on a Dataset stored in a binary file", {
             objective = "regression"
             , metric = "l2"
             , num_leaves = 4L
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , data = dtrain
@@ -504,14 +497,14 @@ test_that("Booster$eval() should work on a Dataset stored in a binary file", {
     eval_from_file <- bst$eval(
         data = lgb.Dataset(
             data = test_file
-            , params = list(verbose = VERBOSITY, num_threads = .LGB_MAX_THREADS)
+            , params = list(verbose = .LGB_VERBOSITY, num_threads = .LGB_MAX_THREADS)
         )$construct()
         , name = "test"
     )
 
-    expect_true(abs(eval_in_mem[[1L]][["value"]] - 0.1744423) < TOLERANCE)
+    expect_true(abs(eval_in_mem[[1L]][["value"]] - 0.1744423) < .LGB_NUMERIC_TOLERANCE)
     # refer to https://github.com/microsoft/LightGBM/issues/4680
-    if (isTRUE(ON_WINDOWS)) {
+    if (isTRUE(.LGB_ON_WINDOWS)) {
       expect_equal(eval_in_mem, eval_from_file)
     } else {
       expect_identical(eval_in_mem, eval_from_file)
@@ -532,7 +525,7 @@ test_that("Booster$rollback_one_iter() should work as expected", {
             num_leaves = 4L
             , learning_rate = 1.0
             , objective = "binary"
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = nrounds
@@ -567,7 +560,7 @@ test_that("Booster$update() passing a train_set works as expected", {
             num_leaves = 4L
             , learning_rate = 1.0
             , objective = "binary"
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = nrounds
@@ -578,7 +571,7 @@ test_that("Booster$update() passing a train_set works as expected", {
         train_set = Dataset$new(
             data = agaricus.train$data
             , label = agaricus.train$label
-            , params = list(verbose = VERBOSITY)
+            , params = list(verbose = .LGB_VERBOSITY)
         )
     )
     expect_true(lgb.is.Booster(bst))
@@ -592,7 +585,7 @@ test_that("Booster$update() passing a train_set works as expected", {
             num_leaves = 4L
             , learning_rate = 1.0
             , objective = "binary"
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = nrounds +  1L
@@ -618,7 +611,7 @@ test_that("Booster$update() throws an informative error if you provide a non-Dat
             num_leaves = 4L
             , learning_rate = 1.0
             , objective = "binary"
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = nrounds
@@ -646,7 +639,7 @@ test_that("Booster should store parameters and Booster$reset_parameter() should
         , metric = c("multi_logloss", "multi_error")
         , boosting = "gbdt"
         , num_class = 5L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     bst <- Booster$new(
@@ -674,7 +667,7 @@ test_that("Booster$params should include dataset params, before and after Booste
         objective = "binary"
         , max_depth = 4L
         , bagging_fraction = 0.8
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     bst <- Booster$new(
@@ -687,7 +680,7 @@ test_that("Booster$params should include dataset params, before and after Booste
             objective = "binary"
             , max_depth = 4L
             , bagging_fraction = 0.8
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
             , max_bin = 17L
         )
@@ -699,7 +692,7 @@ test_that("Booster$params should include dataset params, before and after Booste
         objective = "binary"
         , max_depth = 4L
         , bagging_fraction = 0.9
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
         , max_bin = 17L
     )
@@ -718,7 +711,7 @@ test_that("Saving a model with different feature importance types works", {
             num_leaves = 4L
             , learning_rate = 1.0
             , objective = "binary"
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = 2L
@@ -774,7 +767,7 @@ test_that("Saving a model with unknown importance type fails", {
             num_leaves = 4L
             , learning_rate = 1.0
             , objective = "binary"
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
         )
         , nrounds = 2L
@@ -815,7 +808,7 @@ test_that("all parameters are stored correctly with save_model_to_string()", {
         )
         , data = dtrain
         , nrounds = nrounds
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
     )
 
     model_str <- bst$save_model_to_string()
@@ -832,7 +825,7 @@ test_that("all parameters are stored correctly with save_model_to_string()", {
     expect_equal(sum(params_in_file == "[objective: regression]"), 1L)
 
     expect_equal(sum(startsWith(params_in_file, "[verbosity:")), 1L)
-    expect_equal(sum(params_in_file == sprintf("[verbosity: %i]", VERBOSITY)), 1L)
+    expect_equal(sum(params_in_file == sprintf("[verbosity: %i]", .LGB_VERBOSITY)), 1L)
 
     # early stopping should be off by default
     expect_equal(sum(startsWith(params_in_file, "[early_stopping_round:")), 1L)
@@ -879,7 +872,7 @@ test_that("early_stopping, num_iterations are stored correctly in model string e
         , valids = list(
             "random_valid" = dvalid
         )
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
     )
 
     model_str <- bst$save_model_to_string()
@@ -911,7 +904,7 @@ test_that("Booster: method calls Booster with a null handle should raise an info
             , num_threads = .LGB_MAX_THREADS
         )
         , data = dtrain
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , nrounds = 5L
         , valids = list(
             train = dtrain
@@ -987,7 +980,7 @@ test_that("Booster$new() using a Dataset with a null handle should raise an info
         bst <- Booster$new(
             train_set = dtrain
             , params = list(
-                verbose = VERBOSITY
+                verbose = .LGB_VERBOSITY
             )
         )
     }, regexp = "Attempting to create a Dataset without any raw data")
@@ -1098,7 +1091,7 @@ test_that("lgb.cv() correctly handles passing through params to the model file",
         , n_iter = n_iter
         , early_stopping_round = early_stopping_round
         , n_iter_no_change = n_iter_no_change
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
 
@@ -1108,7 +1101,7 @@ test_that("lgb.cv() correctly handles passing through params to the model file",
         , nrounds = nrounds_kwarg
         , early_stopping_rounds = early_stopping_round_kwarg
         , nfold = 3L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
     )
 
     for (bst in cv_bst$boosters) {
@@ -1143,7 +1136,7 @@ test_that("params (including dataset params) should be stored in .rds file for B
         objective = "binary"
         , max_depth = 4L
         , bagging_fraction = 0.8
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     bst <- Booster$new(
@@ -1160,7 +1153,7 @@ test_that("params (including dataset params) should be stored in .rds file for B
             objective = "binary"
             , max_depth = 4L
             , bagging_fraction = 0.8
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
             , max_bin = 17L
         )
@@ -1180,7 +1173,7 @@ test_that("params (including dataset params) should be stored in .rds file for B
         objective = "binary"
         , max_depth = 4L
         , bagging_fraction = 0.8
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     bst <- Booster$new(
@@ -1197,7 +1190,7 @@ test_that("params (including dataset params) should be stored in .rds file for B
             objective = "binary"
             , max_depth = 4L
             , bagging_fraction = 0.8
-            , verbose = VERBOSITY
+            , verbose = .LGB_VERBOSITY
             , num_threads = .LGB_MAX_THREADS
             , max_bin = 17L
         )
@@ -1212,7 +1205,7 @@ test_that("Handle is automatically restored when calling predict", {
         , nrounds = 5L
         , obj = "binary"
         , params = list(
-            verbose = VERBOSITY
+            verbose = .LGB_VERBOSITY
         )
         , num_threads = .LGB_MAX_THREADS
     )
@@ -1236,7 +1229,7 @@ test_that("boosters with linear models at leaves work with saveRDS.lgb.Booster a
 
     params <- list(
         objective = "regression"
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , metric = "mse"
         , seed = 0L
         , num_leaves = 2L
@@ -1276,7 +1269,7 @@ test_that("boosters with linear models at leaves can be written to RDS and re-lo
 
     params <- list(
         objective = "regression"
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , metric = "mse"
         , seed = 0L
         , num_leaves = 2L
@@ -1386,7 +1379,7 @@ test_that("Booster's print, show, and summary work correctly", {
               min_data_in_bin = 1L
             )
         )
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , nrounds = 5L
     )
     .check_methods_work(model)
@@ -1398,7 +1391,7 @@ test_that("Booster's print, show, and summary work correctly", {
             as.matrix(iris[, -5L])
             , label = as.numeric(factor(iris$Species)) - 1.0
         )
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , nrounds = 5L
     )
     .check_methods_work(model)
@@ -1431,7 +1424,7 @@ test_that("Booster's print, show, and summary work correctly", {
         )
         , obj = .logregobj
         , eval = .evalerror
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , nrounds = 5L
         , params = list(num_threads = .LGB_MAX_THREADS)
     )
@@ -1454,7 +1447,7 @@ test_that("LGBM_BoosterGetNumFeature_R returns correct outputs", {
               min_data_in_bin = 1L
             )
         )
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , nrounds = 5L
     )
     ncols <- .Call(LGBM_BoosterGetNumFeature_R, model$.__enclos_env__$private$handle)
@@ -1467,7 +1460,7 @@ test_that("LGBM_BoosterGetNumFeature_R returns correct outputs", {
             as.matrix(iris[, -5L])
             , label = as.numeric(factor(iris$Species)) - 1.0
         )
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , nrounds = 5L
     )
     ncols <- .Call(LGBM_BoosterGetNumFeature_R, model$.__enclos_env__$private$handle)

R-package/tests/testthat/test_lgb.interprete.R

-VERBOSITY <- as.integer(
-    Sys.getenv("LIGHTGBM_TEST_VERBOSITY", "-1")
-)
-
 .sigmoid <- function(x) {
     1.0 / (1.0 + exp(-x))
 }
@@ -30,7 +26,7 @@ test_that("lgb.intereprete works as expected for binary classification", {
         , max_depth = -1L
         , min_data_in_leaf = 1L
         , min_sum_hessian_in_leaf = 1.0
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     model <- lgb.train(
@@ -83,7 +79,7 @@ test_that("lgb.intereprete works as expected for multiclass classification", {
         , num_class = 3L
         , learning_rate = 0.00001
         , min_data = 1L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     model <- lgb.train(

R-package/tests/testthat/test_lgb.plot.importance.R

-VERBOSITY <- as.integer(
-    Sys.getenv("LIGHTGBM_TEST_VERBOSITY", "-1")
-)
-
 test_that("lgb.plot.importance() should run without error for well-formed inputs", {
     data(agaricus.train, package = "lightgbm")
     train <- agaricus.train
@@ -13,7 +9,7 @@ test_that("lgb.plot.importance() should run without error for well-formed inputs
         , max_depth = -1L
         , min_data_in_leaf = 1L
         , min_sum_hessian_in_leaf = 1.0
-        , verbosity = VERBOSITY
+        , verbosity = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     model <- lgb.train(params, dtrain, 3L)

R-package/tests/testthat/test_lgb.plot.interpretation.R

-VERBOSITY <- as.integer(
-    Sys.getenv("LIGHTGBM_TEST_VERBOSITY", "-1")
-)
-
 .sigmoid <- function(x) {
     1.0 / (1.0 + exp(-x))
 }
@@ -30,7 +26,7 @@ test_that("lgb.plot.interepretation works as expected for binary classification"
         , max_depth = -1L
         , min_data_in_leaf = 1L
         , min_sum_hessian_in_leaf = 1.0
-        , verbosity = VERBOSITY
+        , verbosity = .LGB_VERBOSITY
         , num_threads = .LGB_MAX_THREADS
     )
     model <- lgb.train(
@@ -87,7 +83,7 @@ test_that("lgb.plot.interepretation works as expected for multiclass classificat
         params = params
         , data = dtrain
         , nrounds = 3L
-        , verbose = VERBOSITY
+        , verbose = .LGB_VERBOSITY
     )
     num_trees <- 5L
     tree_interpretation <- lgb.interprete(

R-package/tests/testthat/test_weighted_loss.R

-VERBOSITY <- as.integer(
-  Sys.getenv("LIGHTGBM_TEST_VERBOSITY", "-1")
-)
-
 test_that("Gamma regression reacts on 'weight'", {
   n <- 100L
   set.seed(87L)
@@ -17,7 +13,7 @@ test_that("Gamma regression reacts on 'weight'", {
     params = params
     , data = dtrain
     , nrounds = 4L
-    , verbose = VERBOSITY
+    , verbose = .LGB_VERBOSITY
   )
   pred_unweighted <- predict(bst, X_pred)
 
@@ -31,7 +27,7 @@ test_that("Gamma regression reacts on 'weight'", {
     params = params
     , data = dtrain
     , nrounds = 4L
-    , verbose = VERBOSITY
+    , verbose = .LGB_VERBOSITY
   )
   pred_weighted_1 <- predict(bst, X_pred)
 
@@ -45,7 +41,7 @@ test_that("Gamma regression reacts on 'weight'", {
     params = params
     , data = dtrain
     , nrounds = 4L
-    , verbose = VERBOSITY
+    , verbose = .LGB_VERBOSITY
   )
   pred_weighted_2 <- predict(bst, X_pred)
 
@@ -59,7 +55,7 @@ test_that("Gamma regression reacts on 'weight'", {
     params = params
     , data = dtrain
     , nrounds = 4L
-    , verbose = VERBOSITY
+    , verbose = .LGB_VERBOSITY
   )
   pred_weighted <- predict(bst, X_pred)