[backport] Fix GPU L1 error. (#8749) (#8770)

* [backport] Fix GPU L1 error. (#8749)

* Fix backport.

python-package/xgboost/compat.py

@@ -36,7 +36,6 @@ try:
 
     PANDAS_INSTALLED = True
 except ImportError:
-
     MultiIndex = object
     DataFrame = object
     Series = object
@@ -161,6 +160,7 @@ def concat(value: Sequence[_T]) -> _T:  # pylint: disable=too-many-return-statem
 # `importlib.utils`, except it's unclear from its document on how to use it.  This one
 # seems to be easy to understand and works out of box.
 
+
 # Copyright 2015 The TensorFlow Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License"); you may not use this

src/tree/updater_gpu_hist.cu

@@ -188,7 +188,8 @@ struct GPUHistMakerDevice {
   common::Span<GradientPair> gpair;
 
   dh::device_vector<int> monotone_constraints;
-  dh::device_vector<float> update_predictions;
+  // node idx for each sample
+  dh::device_vector<bst_node_t> positions;
 
   TrainParam param;
 
@@ -426,7 +427,7 @@ struct GPUHistMakerDevice {
         LOG(FATAL) << "Current objective function can not be used with external memory.";
       }
       p_out_position->Resize(0);
-      update_predictions.clear();
+      positions.clear();
       return;
     }
 
@@ -461,8 +462,6 @@ struct GPUHistMakerDevice {
                               HostDeviceVector<bst_node_t>* p_out_position) {
     auto d_matrix = page->GetDeviceAccessor(ctx_->gpu_id);
     auto d_gpair = this->gpair;
-    update_predictions.resize(row_partitioner->GetRows().size());
-    auto d_update_predictions = dh::ToSpan(update_predictions);
     p_out_position->SetDevice(ctx_->gpu_id);
     p_out_position->Resize(row_partitioner->GetRows().size());
 
@@ -497,32 +496,45 @@ struct GPUHistMakerDevice {
         node = d_nodes[position];
       }
 
-      d_update_predictions[row_id] = node.LeafValue();
       return position;
     };  // NOLINT
 
     auto d_out_position = p_out_position->DeviceSpan();
     row_partitioner->FinalisePosition(d_out_position, new_position_op);
 
+    auto s_position = p_out_position->ConstDeviceSpan();
+    positions.resize(s_position.size());
+    dh::safe_cuda(cudaMemcpyAsync(positions.data().get(), s_position.data(),
+                                  s_position.size_bytes(), cudaMemcpyDeviceToDevice));
+
     dh::LaunchN(row_partitioner->GetRows().size(), [=] __device__(size_t idx) {
       bst_node_t position = d_out_position[idx];
-      d_update_predictions[idx] = d_nodes[position].LeafValue();
       bool is_row_sampled = d_gpair[idx].GetHess() - .0f == 0.f;
       d_out_position[idx] = is_row_sampled ? ~position : position;
     });
   }
 
   bool UpdatePredictionCache(linalg::VectorView<float> out_preds_d, RegTree const* p_tree) {
-    if (update_predictions.empty()) {
+    if (positions.empty()) {
       return false;
     }
+
     CHECK(p_tree);
     dh::safe_cuda(cudaSetDevice(ctx_->gpu_id));
     CHECK_EQ(out_preds_d.DeviceIdx(), ctx_->gpu_id);
-    auto d_update_predictions = dh::ToSpan(update_predictions);
-    CHECK_EQ(out_preds_d.Size(), d_update_predictions.size());
-    dh::LaunchN(out_preds_d.Size(), [=] XGBOOST_DEVICE(size_t idx) mutable {
-      out_preds_d(idx) += d_update_predictions[idx];
+
+    auto d_position = dh::ToSpan(positions);
+    CHECK_EQ(out_preds_d.Size(), d_position.size());
+
+    auto const& h_nodes = p_tree->GetNodes();
+    dh::caching_device_vector<RegTree::Node> nodes(h_nodes.size());
+    dh::safe_cuda(cudaMemcpyAsync(nodes.data().get(), h_nodes.data(),
+                                  h_nodes.size() * sizeof(RegTree::Node), cudaMemcpyHostToDevice));
+    auto d_nodes = dh::ToSpan(nodes);
+    dh::LaunchN(d_position.size(), [=] XGBOOST_DEVICE(std::size_t idx) mutable {
+      bst_node_t nidx = d_position[idx];
+      auto weight = d_nodes[nidx].LeafValue();
+      out_preds_d(idx) += weight;
     });
     return true;
   }
@@ -865,6 +877,7 @@ class GPUHistMaker : public TreeUpdater {
   std::unique_ptr<GPUHistMakerDevice<GradientSumT>> maker;  // NOLINT
 
   char const* Name() const override { return "grow_gpu_hist"; }
+  bool HasNodePosition() const override { return true; }
 
  private:
   bool initialised_{false};

tests/cpp/tree/test_node_partition.cc

@@ -0,0 +1,24 @@
+/**
+ * Copyright 2023 by XGBoost contributors
+ */
+#include <gtest/gtest.h>
+#include <xgboost/task.h>
+#include <xgboost/tree_updater.h>
+
+namespace xgboost {
+TEST(Updater, HasNodePosition) {
+  Context ctx;
+  ObjInfo task{ObjInfo::kRegression, true, true};
+  std::unique_ptr<TreeUpdater> up{TreeUpdater::Create("grow_histmaker", &ctx, task)};
+  ASSERT_TRUE(up->HasNodePosition());
+
+  up.reset(TreeUpdater::Create("grow_quantile_histmaker", &ctx, task));
+  ASSERT_TRUE(up->HasNodePosition());
+
+#if defined(XGBOOST_USE_CUDA)
+  ctx.gpu_id = 0;
+  up.reset(TreeUpdater::Create("grow_gpu_hist", &ctx, task));
+  ASSERT_TRUE(up->HasNodePosition());
+#endif  // defined(XGBOOST_USE_CUDA)
+}
+}  // namespace xgboost

tests/python-gpu/test_gpu_prediction.py

@@ -338,13 +338,21 @@ class TestGPUPredict:
     @given(predict_parameter_strategy, tm.dataset_strategy)
     @settings(deadline=None, max_examples=20, print_blob=True)
     def test_predict_leaf_gbtree(self, param, dataset):
+        # Unsupported for random forest
+        if param.get("num_parallel_tree", 1) > 1 and dataset.name.endswith("-l1"):
+            return
+
         param['booster'] = 'gbtree'
         param['tree_method'] = 'gpu_hist'
         self.run_predict_leaf_booster(param, 10, dataset)
 
     @given(predict_parameter_strategy, tm.dataset_strategy)
     @settings(deadline=None, max_examples=20, print_blob=True)
-    def test_predict_leaf_dart(self, param, dataset):
+    def test_predict_leaf_dart(self, param: dict, dataset: tm.TestDataset) -> None:
+        # Unsupported for random forest
+        if param.get("num_parallel_tree", 1) > 1 and dataset.name.endswith("-l1"):
+            return
+
         param['booster'] = 'dart'
         param['tree_method'] = 'gpu_hist'
         self.run_predict_leaf_booster(param, 10, dataset)

tests/python/test_updaters.py

@@ -458,6 +458,22 @@ class TestTreeMethod:
         config_0 = json.loads(booster_0.save_config())
         np.testing.assert_allclose(get_score(config_0), get_score(config_1) + 1)
 
+        evals_result: Dict[str, Dict[str, list]] = {}
+        xgb.train(
+            {
+                "tree_method": tree_method,
+                "objective": "reg:absoluteerror",
+                "subsample": 0.8
+            },
+            Xy,
+            num_boost_round=10,
+            evals=[(Xy, "Train")],
+            evals_result=evals_result,
+        )
+        mae = evals_result["Train"]["mae"]
+        assert mae[-1] < 20.0
+        assert tm.non_increasing(mae)
+
     @pytest.mark.skipif(**tm.no_sklearn())
     @pytest.mark.parametrize(
         "tree_method,weighted", [