[bp] Fix rng for the column sampler. (dmlc#10998) (dmlc#11004)

Author: trivialfis

src/common/random.h

@@ -230,7 +230,7 @@ class ColumnSampler {
 };
 
 inline auto MakeColumnSampler(Context const* ctx) {
-  std::uint32_t seed = common::GlobalRandomEngine()();
+  std::uint32_t seed = common::GlobalRandom()();
   auto rc = collective::Broadcast(ctx, linalg::MakeVec(&seed, 1), 0);
   collective::SafeColl(rc);
   auto cs = std::make_shared<common::ColumnSampler>(seed);

src/tree/updater_gpu_hist.cu

@@ -867,12 +867,7 @@ class GPUHistMaker : public TreeUpdater {
     CHECK_GE(ctx_->Ordinal(), 0) << "Must have at least one device";
     info_ = &dmat->Info();
 
-    // Synchronise the column sampling seed
-    uint32_t column_sampling_seed = common::GlobalRandom()();
-    auto rc = collective::Broadcast(
-        ctx_, linalg::MakeVec(&column_sampling_seed, sizeof(column_sampling_seed)), 0);
-    SafeColl(rc);
-    this->column_sampler_ = std::make_shared<common::ColumnSampler>(column_sampling_seed);
+    this->column_sampler_ = common::MakeColumnSampler(ctx_);
 
     auto batch_param = BatchParam{param->max_bin, TrainParam::DftSparseThreshold()};
     dh::safe_cuda(cudaSetDevice(ctx_->Ordinal()));
@@ -1012,8 +1007,7 @@ class GPUGlobalApproxMaker : public TreeUpdater {
 
     monitor_.Start(__func__);
     CHECK(ctx_->IsCUDA()) << error::InvalidCUDAOrdinal();
-    uint32_t column_sampling_seed = common::GlobalRandom()();
-    this->column_sampler_ = std::make_shared<common::ColumnSampler>(column_sampling_seed);
+    this->column_sampler_ = common::MakeColumnSampler(ctx_);
 
     p_last_fmat_ = p_fmat;
     initialised_ = true;

tests/python/test_updaters.py

@@ -54,6 +54,28 @@ def test_exact_sample_by_node_error(self) -> None:
             num_boost_round=2,
         )
 
+    @pytest.mark.parametrize("tree_method", ["approx", "hist"])
+    def test_colsample_rng(self, tree_method: str) -> None:
+        """Test rng has an effect on column sampling."""
+        X, y, _ = tm.make_regression(128, 16, use_cupy=False)
+        reg0 = xgb.XGBRegressor(
+            n_estimators=2,
+            colsample_bynode=0.5,
+            random_state=42,
+            tree_method=tree_method,
+        )
+        reg0.fit(X, y)
+
+        reg1 = xgb.XGBRegressor(
+            n_estimators=2,
+            colsample_bynode=0.5,
+            random_state=43,
+            tree_method=tree_method,
+        )
+        reg1.fit(X, y)
+
+        assert list(reg0.feature_importances_) != list(reg1.feature_importances_)
+
     @given(
         exact_parameter_strategy,
         hist_parameter_strategy,