Hendrik/xgboost
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Prepare external memory support for hist. (#7638)

Browse Source 
This PR prepares the GHistIndexMatrix to host the column matrix which is used by the hist tree method by accepting sparse_threshold parameter.

Some cleanups are made to ensure the correct batch param is being passed into DMatrix along with some additional tests for correctness of SimpleDMatrix.
This commit is contained in:
Jiaming Yuan
2022-02-10 16:58:02 +08:00
committed by GitHub
parent 87c01f49d8
commit 2775c2a1ab
24 changed files with 368 additions and 201 deletions
Download Patch File Download Diff File Expand all files Collapse all files

41
tests/cpp/tree/test_approx.cc
View File

@@ -8,6 +8,19 @@
namespace xgboost {
namespace tree {
namespace {
void GetSplit(RegTree *tree, float split_value, std::vector<CPUExpandEntry> *candidates) {
tree->ExpandNode(
/*nid=*/RegTree::kRoot, /*split_index=*/0, /*split_value=*/split_value,
/*default_left=*/true, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f,
/*left_sum=*/0.0f,
/*right_sum=*/0.0f);
candidates->front().split.split_value = split_value;
candidates->front().split.sindex = 0;
candidates->front().split.sindex |= (1U << 31);
}
} // anonymous namespace
TEST(Approx, Partitioner) {
size_t n_samples = 1024, n_features = 1, base_rowid = 0;
ApproxRowPartitioner partitioner{n_samples, base_rowid};
@@ -20,20 +33,18 @@ TEST(Approx, Partitioner) {
ctx.InitAllowUnknown(Args{});
std::vector<CPUExpandEntry> candidates{{0, 0, 0.4}};
for (auto const &page : Xy->GetBatches<GHistIndexMatrix>({GenericParameter::kCpuId, 64})) {
bst_feature_t split_ind = 0;
auto grad = GenerateRandomGradients(n_samples);
std::vector<float> hess(grad.Size());
std::transform(grad.HostVector().cbegin(), grad.HostVector().cend(), hess.begin(),
[](auto gpair) { return gpair.GetHess(); });
for (auto const &page : Xy->GetBatches<GHistIndexMatrix>({64, hess, true})) {
bst_feature_t const split_ind = 0;
{
auto min_value = page.cut.MinValues()[split_ind];
RegTree tree;
tree.ExpandNode(
/*nid=*/0, /*split_index=*/0, /*split_value=*/min_value,
/*default_left=*/true, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f,
/*left_sum=*/0.0f,
/*right_sum=*/0.0f);
ApproxRowPartitioner partitioner{n_samples, base_rowid};
candidates.front().split.split_value = min_value;
candidates.front().split.sindex = 0;
candidates.front().split.sindex |= (1U << 31);
GetSplit(&tree, min_value, &candidates);
partitioner.UpdatePosition(&ctx, page, candidates, &tree);
ASSERT_EQ(partitioner.Size(), 3);
ASSERT_EQ(partitioner[1].Size(), 0);
@@ -44,16 +55,8 @@ TEST(Approx, Partitioner) {
auto ptr = page.cut.Ptrs()[split_ind + 1];
float split_value = page.cut.Values().at(ptr / 2);
RegTree tree;
tree.ExpandNode(
/*nid=*/RegTree::kRoot, /*split_index=*/split_ind,
/*split_value=*/split_value,
/*default_left=*/true, 0.0f, 0.0f, 0.0f, 0.0f, 0.0f,
/*left_sum=*/0.0f,
/*right_sum=*/0.0f);
GetSplit(&tree, split_value, &candidates);
auto left_nidx = tree[RegTree::kRoot].LeftChild();
candidates.front().split.split_value = split_value;
candidates.front().split.sindex = 0;
candidates.front().split.sindex |= (1U << 31);
partitioner.UpdatePosition(&ctx, page, candidates, &tree);
auto elem = partitioner[left_nidx];