[backport] Fix CPU bin compression with categorical data. (#8809) (#8810)

* [backport] Fix CPU bin compression with categorical data. (#8809)

* Fix CPU bin compression with categorical data.

* The bug causes the maximum category to be lesser than 256 or the maximum number of bins when
the input data is dense.

* Avoid test symbol.

src/common/column_matrix.cc

@@ -46,7 +46,7 @@ void ColumnMatrix::InitStorage(GHistIndexMatrix const& gmat, double sparse_thres
     feature_offsets_[fid] = accum_index;
   }
 
-  SetTypeSize(gmat.max_num_bins);
+  SetTypeSize(gmat.MaxNumBinPerFeat());
   auto storage_size =
       feature_offsets_.back() * static_cast<std::underlying_type_t<BinTypeSize>>(bins_type_size_);
   index_.resize(storage_size, 0);

src/data/gradient_index.cc

@@ -20,13 +20,13 @@ GHistIndexMatrix::GHistIndexMatrix() : columns_{std::make_unique<common::ColumnM
 
 GHistIndexMatrix::GHistIndexMatrix(DMatrix *p_fmat, bst_bin_t max_bins_per_feat,
                                    double sparse_thresh, bool sorted_sketch, int32_t n_threads,
-                                   common::Span<float> hess) {
+                                   common::Span<float> hess)
+    : max_numeric_bins_per_feat{max_bins_per_feat} {
   CHECK(p_fmat->SingleColBlock());
   // We use sorted sketching for approx tree method since it's more efficient in
   // computation time (but higher memory usage).
   cut = common::SketchOnDMatrix(p_fmat, max_bins_per_feat, n_threads, sorted_sketch, hess);
 
-  max_num_bins = max_bins_per_feat;
   const uint32_t nbins = cut.Ptrs().back();
   hit_count.resize(nbins, 0);
   hit_count_tloc_.resize(n_threads * nbins, 0);
@@ -63,7 +63,7 @@ GHistIndexMatrix::GHistIndexMatrix(MetaInfo const &info, common::HistogramCuts &
     : row_ptr(info.num_row_ + 1, 0),
       hit_count(cuts.TotalBins(), 0),
       cut{std::forward<common::HistogramCuts>(cuts)},
-      max_num_bins(max_bin_per_feat),
+      max_numeric_bins_per_feat(max_bin_per_feat),
       isDense_{info.num_col_ * info.num_row_ == info.num_nonzero_} {}
 
 #if !defined(XGBOOST_USE_CUDA)
@@ -86,13 +86,13 @@ void GHistIndexMatrix::PushBatch(SparsePage const &batch, common::Span<FeatureTy
 }
 
 GHistIndexMatrix::GHistIndexMatrix(SparsePage const &batch, common::Span<FeatureType const> ft,
-                                   common::HistogramCuts const &cuts, int32_t max_bins_per_feat,
+                                   common::HistogramCuts cuts, int32_t max_bins_per_feat,
-                                   bool isDense, double sparse_thresh, int32_t n_threads) {
+                                   bool isDense, double sparse_thresh, int32_t n_threads)
+    : cut{std::move(cuts)},
+      max_numeric_bins_per_feat{max_bins_per_feat},
+      base_rowid{batch.base_rowid},
+      isDense_{isDense} {
   CHECK_GE(n_threads, 1);
-  base_rowid = batch.base_rowid;
-  isDense_ = isDense;
-  cut = cuts;
-  max_num_bins = max_bins_per_feat;
   CHECK_EQ(row_ptr.size(), 0);
   // The number of threads is pegged to the batch size. If the OMP
   // block is parallelized on anything other than the batch/block size,
@@ -127,12 +127,13 @@ INSTANTIATION_PUSH(data::SparsePageAdapterBatch)
 #undef INSTANTIATION_PUSH
 
 void GHistIndexMatrix::ResizeIndex(const size_t n_index, const bool isDense) {
-  if ((max_num_bins - 1 <= static_cast<int>(std::numeric_limits<uint8_t>::max())) && isDense) {
+  if ((MaxNumBinPerFeat() - 1 <= static_cast<int>(std::numeric_limits<uint8_t>::max())) &&
+      isDense) {
     // compress dense index to uint8
     index.SetBinTypeSize(common::kUint8BinsTypeSize);
     index.Resize((sizeof(uint8_t)) * n_index);
-  } else if ((max_num_bins - 1 > static_cast<int>(std::numeric_limits<uint8_t>::max()) &&
+  } else if ((MaxNumBinPerFeat() - 1 > static_cast<int>(std::numeric_limits<uint8_t>::max()) &&
-              max_num_bins - 1 <= static_cast<int>(std::numeric_limits<uint16_t>::max())) &&
+              MaxNumBinPerFeat() - 1 <= static_cast<int>(std::numeric_limits<uint16_t>::max())) &&
              isDense) {
     // compress dense index to uint16
     index.SetBinTypeSize(common::kUint16BinsTypeSize);

src/data/gradient_index.cu

@@ -65,7 +65,7 @@ void GetRowPtrFromEllpack(Context const* ctx, EllpackPageImpl const* page,
 
 GHistIndexMatrix::GHistIndexMatrix(Context const* ctx, MetaInfo const& info,
                                    EllpackPage const& in_page, BatchParam const& p)
-    : max_num_bins{p.max_bin} {
+    : max_numeric_bins_per_feat{p.max_bin} {
   auto page = in_page.Impl();
   isDense_ = page->is_dense;
 

src/data/gradient_index.h

@@ -133,11 +133,15 @@ class GHistIndexMatrix {
   std::vector<size_t> hit_count;
   /*! \brief The corresponding cuts */
   common::HistogramCuts cut;
-  /*! \brief max_bin for each feature. */
+  /** \brief max_bin for each feature. */
-  bst_bin_t max_num_bins;
+  bst_bin_t max_numeric_bins_per_feat;
   /*! \brief base row index for current page (used by external memory) */
   size_t base_rowid{0};
 
+  bst_bin_t MaxNumBinPerFeat() const {
+    return std::max(static_cast<bst_bin_t>(cut.MaxCategory() + 1), max_numeric_bins_per_feat);
+  }
+
   ~GHistIndexMatrix();
   /**
    * \brief Constrcutor for SimpleDMatrix.
@@ -160,7 +164,7 @@ class GHistIndexMatrix {
    * \brief Constructor for external memory.
    */
   GHistIndexMatrix(SparsePage const& page, common::Span<FeatureType const> ft,
-                   common::HistogramCuts const& cuts, int32_t max_bins_per_feat, bool is_dense,
+                   common::HistogramCuts cuts, int32_t max_bins_per_feat, bool is_dense,
                    double sparse_thresh, int32_t n_threads);
   GHistIndexMatrix();  // also for ext mem, empty ctor so that we can read the cache back.
 

src/data/gradient_index_format.cc

@@ -35,7 +35,7 @@ class GHistIndexRawFormat : public SparsePageFormat<GHistIndexMatrix> {
     if (!fi->Read(&page->hit_count)) {
       return false;
     }
-    if (!fi->Read(&page->max_num_bins)) {
+    if (!fi->Read(&page->max_numeric_bins_per_feat)) {
       return false;
     }
     if (!fi->Read(&page->base_rowid)) {
@@ -76,8 +76,8 @@ class GHistIndexRawFormat : public SparsePageFormat<GHistIndexMatrix> {
         page.hit_count.size() * sizeof(decltype(page.hit_count)::value_type) +
         sizeof(uint64_t);
     // max_bins, base row, is_dense
-    fo->Write(page.max_num_bins);
+    fo->Write(page.max_numeric_bins_per_feat);
-    bytes += sizeof(page.max_num_bins);
+    bytes += sizeof(page.max_numeric_bins_per_feat);
     fo->Write(page.base_rowid);
     bytes += sizeof(page.base_rowid);
     fo->Write(page.IsDense());

tests/cpp/data/test_gradient_index.cc

@@ -68,6 +68,30 @@ TEST(GradientIndex, FromCategoricalBasic) {
   }
 }
 
+TEST(GradientIndex, FromCategoricalLarge) {
+  size_t constexpr kRows = 1000, kCats = 512, kCols = 1;
+  bst_bin_t max_bins = 8;
+  auto x = GenerateRandomCategoricalSingleColumn(kRows, kCats);
+  auto m = GetDMatrixFromData(x, kRows, 1);
+  Context ctx;
+
+  auto &h_ft = m->Info().feature_types.HostVector();
+  h_ft.resize(kCols, FeatureType::kCategorical);
+
+  BatchParam p{max_bins, 0.8};
+  {
+    GHistIndexMatrix gidx(m.get(), max_bins, p.sparse_thresh, false, Context{}.Threads(), {});
+    ASSERT_TRUE(gidx.index.GetBinTypeSize() == common::kUint16BinsTypeSize);
+  }
+  {
+    for (auto const &page : m->GetBatches<GHistIndexMatrix>(p)) {
+      common::HistogramCuts cut = page.cut;
+      GHistIndexMatrix gidx{m->Info(), std::move(cut), max_bins};
+      ASSERT_EQ(gidx.MaxNumBinPerFeat(), kCats);
+    }
+  }
+}
+
 TEST(GradientIndex, PushBatch) {
   size_t constexpr kRows = 64, kCols = 4;
   bst_bin_t max_bins = 64;