Fix spelling in documents (#6948)

* Update roxygen2 doc.

Co-authored-by: fis <jm.yuan@outlook.com>

src/c_api/c_api.cc

@@ -638,10 +638,10 @@ XGB_DLL int XGBoosterPredictFromDMatrix(BoosterHandle handle,
                                         bst_float const **out_result) {
   API_BEGIN();
   if (handle == nullptr) {
-    LOG(FATAL) << "Booster has not been intialized or has already been disposed.";
+    LOG(FATAL) << "Booster has not been initialized or has already been disposed.";
   }
   if (dmat == nullptr) {
-    LOG(FATAL) << "DMatrix has not been intialized or has already been disposed.";
+    LOG(FATAL) << "DMatrix has not been initialized or has already been disposed.";
   }
   auto config = Json::Load(StringView{c_json_config});
 

src/c_api/c_api_error.h

@@ -36,7 +36,7 @@
   return 0; // NOLINT(*)
 
 #define CHECK_HANDLE() if (handle == nullptr) \
-  LOG(FATAL) << "DMatrix/Booster has not been intialized or has already been disposed.";
+  LOG(FATAL) << "DMatrix/Booster has not been initialized or has already been disposed.";
 
 /*!
  * \brief Set the last error message needed by C API

src/common/charconv.cc

@@ -1,7 +1,7 @@
 /*!
  * Copyright 2020 by XGBoost Contributors
  *
- * \brief An implemenation of Ryu algorithm:
+ * \brief An implementation of Ryu algorithm:
  *
  * https://dl.acm.org/citation.cfm?id=3192369
  *
@@ -686,7 +686,7 @@ int32_t ToCharsFloatImpl(float f, char * const result) {
 
 // This is an implementation for base 10 inspired by the one in libstdc++v3.  The general
 // scheme is by decomposing the value into multiple combination of base (which is 10) by
-// mod, until the value is lesser than 10, then last char is just char '0' (ascii 48) plus
+// mod, until the value is lesser than 10, then last char is just char '0' (ASCII 48) plus
 // that value.  Other popular implementations can be found in RapidJson and libc++ (in
 // llvm-project), which uses the same general work flow with the same look up table, but
 // probably with better performance as they are more complicated.

src/common/common.h

@@ -55,7 +55,7 @@ namespace xgboost {
 namespace common {
 /*!
  * \brief Split a string by delimiter
- * \param s String to be splitted.
+ * \param s String to be split.
  * \param delim The delimiter.
  */
 inline std::vector<std::string> Split(const std::string& s, char delim) {

src/common/device_helpers.cuh

@@ -253,7 +253,7 @@ __global__ void LaunchNKernel(int device_idx, size_t begin, size_t end,
  *   function as argument.  Hence functions like `LaunchN` cannot use this wrapper.
  *
  * - With c++ initialization list `{}` syntax, you are forced to comply with the CUDA type
- *   spcification.
+ *   specification.
  */
 class LaunchKernel {
   size_t shmem_size_;
@@ -930,7 +930,7 @@ class SegmentSorter {
   // Items sorted within the group
   caching_device_vector<T> ditems_;
 
-  // Original position of the items before they are sorted descendingly within its groups
+  // Original position of the items before they are sorted descending within their groups
   caching_device_vector<uint32_t> doriginal_pos_;
 
   // Segments within the original list that delineates the different groups

src/common/hist_util.h

@@ -81,7 +81,7 @@ class HistogramCuts {
   }
 
   // Getters.  Cuts should be of no use after building histogram indices, but currently
-  // it's deeply linked with quantile_hist, gpu sketcher and gpu_hist.  So we preserve
+  // they are deeply linked with quantile_hist, gpu sketcher and gpu_hist, so we preserve
   // these for now.
   std::vector<uint32_t> const& Ptrs()      const { return cut_ptrs_.ConstHostVector();   }
   std::vector<float>    const& Values()    const { return cut_values_.ConstHostVector(); }
@@ -247,7 +247,7 @@ struct GHistIndexMatrix {
   // Create a global histogram matrix, given cut
   void Init(DMatrix* p_fmat, int max_num_bins);
 
-  // specific method for sparse data as no posibility to reduce allocated memory
+  // specific method for sparse data as no possibility to reduce allocated memory
   template <typename BinIdxType, typename GetOffset>
   void SetIndexData(common::Span<BinIdxType> index_data_span,
                     size_t batch_threads, const SparsePage &batch,

src/common/json.cc

@@ -394,7 +394,7 @@ Json JsonReader::Parse() {
       return ParseArray();
     } else if ( c == '-' || std::isdigit(c) ||
                 c == 'N' || c == 'I') {
-      // For now we only accept `NaN`, not `nan` as the later violiates LR(1) with `null`.
+      // For now we only accept `NaN`, not `nan` as the later violates LR(1) with `null`.
       return ParseNumber();
     } else if ( c == '\"' ) {
       return ParseString();

src/common/math.h

@@ -77,7 +77,7 @@ XGBOOST_DEVICE inline void Softmax(Iterator start, Iterator end) {
 
 /*!
  * \brief Find the maximum iterator within the iterators
- * \param begin The begining iterator.
+ * \param begin The beginning iterator.
  * \param end The end iterator.
  * \return the iterator point to the maximum value.
  * \tparam Iterator The type of the iterator.
@@ -107,7 +107,7 @@ inline float LogSum(float x, float y) {
 
 /*!
  * \brief perform numerically safe logsum
- * \param begin The begining iterator.
+ * \param begin The beginning iterator.
  * \param end The end iterator.
  * \return the iterator point to the maximum value.
  * \tparam Iterator The type of the iterator.
@@ -135,7 +135,7 @@ inline static bool CmpSecond(const std::pair<float, unsigned> &a,
   return a.second > b.second;
 }
 
-// Redefined here to workaround a VC bug that doesn't support overloadng for integer
+// Redefined here to workaround a VC bug that doesn't support overloading for integer
 // types.
 template <typename T>
 XGBOOST_DEVICE typename std::enable_if<

src/common/quantile.cc

@@ -55,7 +55,7 @@ HostSketchContainer::CalcColumnSize(SparsePage const &batch,
 std::vector<bst_feature_t> HostSketchContainer::LoadBalance(
     SparsePage const &batch, bst_feature_t n_columns, size_t const nthreads) {
   /* Some sparse datasets have their mass concentrating on small number of features.  To
-   * avoid wating for a few threads running forever, we here distirbute different number
+   * avoid waiting for a few threads running forever, we here distribute different number
    * of columns to different threads according to number of entries.
    */
   auto page = batch.GetView();

src/common/quantile.cu

@@ -184,9 +184,9 @@ common::Span<thrust::tuple<uint64_t, uint64_t>> MergePath(
       });
 
   // Compute the index for both x and y (which of the element in a and b are used in each
-  // comparison) by scaning the binary merge path.  Take output [(x_0, y_0), (x_0, y_1),
+  // comparison) by scanning the binary merge path.  Take output [(x_0, y_0), (x_0, y_1),
   // ...] as an example, the comparison between (x_0, y_0) adds 1 step in the merge path.
-  // Asumming y_0 is less than x_0 so this step is torward the end of y.  After the
+  // Assuming y_0 is less than x_0 so this step is toward the end of y.  After the
   // comparison, index of y is incremented by 1 from y_0 to y_1, and at the same time, y_0
   // is landed into output as the first element in merge result.  The scan result is the
   // subscript of x and y.
@@ -367,7 +367,7 @@ void SketchContainer::Push(Span<Entry const> entries, Span<size_t> columns_ptr,
 size_t SketchContainer::ScanInput(Span<SketchEntry> entries, Span<OffsetT> d_columns_ptr_in) {
   /* There are 2 types of duplication.  First is duplicated feature values, which comes
    * from user input data.  Second is duplicated sketching entries, which is generated by
-   * prunning or merging. We preserve the first type and remove the second type.
+   * pruning or merging. We preserve the first type and remove the second type.
    */
   timer_.Start(__func__);
   dh::safe_cuda(cudaSetDevice(device_));

src/common/survival_util.h

@@ -44,7 +44,7 @@ constexpr double kMaxGradient = 15.0;
 constexpr double kMinHessian = 1e-16;  // Ensure that no data point gets zero hessian
 constexpr double kMaxHessian = 15.0;
 
-constexpr double kEps = 1e-12;  // A denomitor in a fraction should not be too small
+constexpr double kEps = 1e-12;  // A denominator in a fraction should not be too small
 
 // Clip (limit) x to fit range [x_min, x_max].
 // If x < x_min, return x_min; if x > x_max, return x_max; if x_min <= x <= x_max, return x.

src/common/transform.h

@@ -52,7 +52,7 @@ __global__ void LaunchCUDAKernel(Functor _func, Range _range,
  *
  *     If you use it in a function that can be compiled by both nvcc and host
  *     compiler, the behaviour is un-defined!  Because your function is NOT
- *     duplicated by `CompiledWithCuda`. At link time, cuda compiler resolution
+ *     duplicated by `CompiledWithCuda`. At link time, CUDA compiler resolution
  *     will merge functions with same signature.
  */
 template <bool CompiledWithCuda = WITH_CUDA()>
@@ -155,7 +155,7 @@ class Transform {
           _func, shard_range, UnpackHDVOnDevice(_vectors)...);
     }
 #else
-    /*! \brief Dummy funtion defined when compiling for CPU.  */
+    /*! \brief Dummy function defined when compiling for CPU.  */
     template <typename std::enable_if<!CompiledWithCuda>::type* = nullptr,
               typename... HDV>
     void LaunchCUDA(Functor _func, HDV*...) const {

src/common/version.cc

@@ -36,7 +36,7 @@ Version::TripletT Version::Load(Json const& in) {
 
 Version::TripletT Version::Load(dmlc::Stream* fi) {
   XGBoostVersionT major{0}, minor{0}, patch{0};
-  // This is only used in DMatrix serialization, so doesn't break model compability.
+  // This is only used in DMatrix serialization, so doesn't break model compatibility.
   std::string msg { "Incorrect version format found in binary file.  "
                     "Binary file from XGBoost < 1.0.0 is no longer supported. "
                     "Please generate it again." };

src/common/version.h

@@ -17,7 +17,7 @@ struct Version {
   using TripletT = std::tuple<XGBoostVersionT, XGBoostVersionT, XGBoostVersionT>;
   static const TripletT kInvalid;
 
-  // Save/Load version info to Json document
+  // Save/Load version info to JSON document
   static TripletT Load(Json const& in);
   static void Save(Json* out);
 

src/data/adapter.h

@@ -42,7 +42,7 @@ namespace data {
  * This abstraction allows us to read through different sparse matrix formats
  * using the same interface. In particular we can write a DMatrix constructor
  * that uses the same code to construct itself from a CSR matrix, CSC matrix,
- * dense matrix, csv, libsvm file, or potentially other formats. To see why this
+ * dense matrix, CSV, LIBSVM file, or potentially other formats. To see why this
  * is necessary, imagine we have 5 external matrix formats and 5 internal
  * DMatrix types where each DMatrix needs a custom constructor for each possible
  * input. The number of constructors is 5*5=25. Using an abstraction over the
@@ -736,7 +736,7 @@ class IteratorAdapter : public dmlc::DataIter<FileAdapterBatch> {
 
   size_t columns_;
   size_t row_offset_;
-  // at the beinning.
+  // at the beginning.
   bool at_first_;
   // handle to the iterator,
   DataIterHandle data_handle_;

src/data/ellpack_page.cu

@@ -187,7 +187,7 @@ template <typename AdapterBatchT>
 void CopyDataToEllpack(const AdapterBatchT& batch, EllpackPageImpl* dst,
                        int device_idx, float missing) {
   // Some witchcraft happens here
-  // The goal is to copy valid elements out of the input to an ellpack matrix
+  // The goal is to copy valid elements out of the input to an ELLPACK matrix
   // with a given row stride, using no extra working memory Standard stream
   // compaction needs to be modified to do this, so we manually define a
   // segmented stream compaction via operators on an inclusive scan. The output

src/data/ellpack_page.cuh

@@ -13,13 +13,13 @@
 #include <thrust/binary_search.h>
 
 namespace xgboost {
-/** \brief Struct for accessing and manipulating an ellpack matrix on the
+/** \brief Struct for accessing and manipulating an ELLPACK matrix on the
  * device. Does not own underlying memory and may be trivially copied into
  * kernels.*/
 struct EllpackDeviceAccessor {
   /*! \brief Whether or not if the matrix is dense. */
   bool is_dense;
-  /*! \brief Row length for ELLPack, equal to number of features. */
+  /*! \brief Row length for ELLPACK, equal to number of features. */
   size_t row_stride;
   size_t base_rowid{};
   size_t n_rows{};
@@ -197,11 +197,11 @@ class EllpackPageImpl {
 public:
   /*! \brief Whether or not if the matrix is dense. */
   bool is_dense;
-  /*! \brief Row length for ELLPack. */
+  /*! \brief Row length for ELLPACK. */
   size_t row_stride;
   size_t base_rowid{0};
   size_t n_rows{};
-  /*! \brief global index of histogram, which is stored in ELLPack format. */
+  /*! \brief global index of histogram, which is stored in ELLPACK format. */
   HostDeviceVector<common::CompressedByteT> gidx_buffer;
 
  private:

src/gbm/gbtree.cc

@@ -563,7 +563,7 @@ GBTree::GetPredictor(HostDeviceVector<float> const *out_pred,
   // GPU_Hist by default has prediction cache calculated from quantile values,
   // so GPU Predictor is not used for training dataset.  But when XGBoost
   // performs continue training with an existing model, the prediction cache is
-  // not availbale and number of trees doesn't equal zero, the whole training
+  // not available and number of trees doesn't equal zero, the whole training
   // dataset got copied into GPU for precise prediction.  This condition tries
   // to avoid such copy by calling CPU Predictor instead.
   if ((out_pred && out_pred->Size() == 0) && (model_.param.num_trees != 0) &&
@@ -831,7 +831,7 @@ class Dart : public GBTree {
 #pragma omp parallel for
         for (omp_ulong ridx = 0; ridx < n_rows; ++ridx) {
           const size_t offset = ridx * n_groups + group;
-          // Need to remove the base margin from indiviual tree.
+          // Need to remove the base margin from individual tree.
           h_out_predts[offset] +=
               (h_predts[offset] - model_.learner_model_param->base_score) * w;
         }

src/metric/rank_metric.cc

@@ -5,9 +5,9 @@
 // possible for a valid device ordinal to be present for non GPU builds. However, it is possible
 // for an invalid device ordinal to be specified in GPU builds - to train/predict and/or compute
 // the metrics on CPU. To accommodate these scenarios, the following is done for the metrics
-// accelarated on the GPU.
+// accelerated on the GPU.
 // - An internal GPU registry holds all the GPU metric types (defined in the .cu file)
-// - An instance of the appropriate gpu metric type is created when a device ordinal is present
+// - An instance of the appropriate GPU metric type is created when a device ordinal is present
 // - If the creation is successful, the metric computation is done on the device
 // - else, it falls back on the CPU
 // - The GPU metric types are *only* registered when xgboost is built for GPUs
@@ -561,7 +561,7 @@ XGBOOST_REGISTER_METRIC(MAP, "map")
 .set_body([](const char* param) { return new EvalMAP("map", param); });
 
 XGBOOST_REGISTER_METRIC(Cox, "cox-nloglik")
-.describe("Negative log partial likelihood of Cox proportioanl hazards model.")
+.describe("Negative log partial likelihood of Cox proportional hazards model.")
 .set_body([](const char*) { return new EvalCox(); });
 }  // namespace metric
 }  // namespace xgboost

src/objective/regression_obj.cu

@@ -271,7 +271,7 @@ class PoissonRegression : public ObjFunction {
 DMLC_REGISTER_PARAMETER(PoissonRegressionParam);
 
 XGBOOST_REGISTER_OBJECTIVE(PoissonRegression, "count:poisson")
-.describe("Possion regression for count data.")
+.describe("Poisson regression for count data.")
 .set_body([]() { return new PoissonRegression(); });
 
 

src/tree/constraints.h

@@ -17,7 +17,7 @@ namespace xgboost {
 /*!
  * \brief Feature interaction constraint implementation for CPU tree updaters.
  *
- * The interface is similiar to the one for GPU Hist.
+ * The interface is similar to the one for GPU Hist.
  */
 class FeatureInteractionConstraintHost {
  protected:

src/tree/gpu_hist/evaluate_splits.cu

@@ -125,7 +125,7 @@ struct UpdateNumeric {
                              EvaluateSplitInputs<GradientSumT> const &inputs,
                              DeviceSplitCandidate *best_split) {
     // Use pointer from cut to indicate begin and end of bins for each feature.
-    uint32_t gidx_begin = inputs.feature_segments[fidx];  // begining bin
+    uint32_t gidx_begin = inputs.feature_segments[fidx];  // beginning bin
     int split_gidx = (scan_begin + threadIdx.x) - 1;
     float fvalue;
     if (split_gidx < static_cast<int>(gidx_begin)) {
@@ -152,7 +152,7 @@ __device__ void EvaluateFeature(
     TempStorageT* temp_storage         // temp memory for cub operations
 ) {
   // Use pointer from cut to indicate begin and end of bins for each feature.
-  uint32_t gidx_begin = inputs.feature_segments[fidx];  // begining bin
+  uint32_t gidx_begin = inputs.feature_segments[fidx];  // beginning bin
   uint32_t gidx_end =
       inputs.feature_segments[fidx + 1];  // end bin for i^th feature
   auto feature_hist = inputs.gradient_histogram.subspan(gidx_begin, gidx_end - gidx_begin);

src/tree/gpu_hist/gradient_based_sampler.cuh

@@ -124,7 +124,7 @@ class ExternalMemoryGradientBasedSampling : public SamplingStrategy {
  * Processing Systems (pp. 3146-3154).
  * \see Zhu, R. (2016). Gradient-based sampling: An adaptive importance sampling for least-squares.
  * In Advances in Neural Information Processing Systems (pp. 406-414).
- * \see Ohlsson, E. (1998). Sequential poisson sampling. Journal of official Statistics, 14(2), 149.
+ * \see Ohlsson, E. (1998). Sequential Poisson sampling. Journal of official Statistics, 14(2), 149.
  */
 class GradientBasedSampler {
  public:

src/tree/gpu_hist/histogram.cu

@@ -17,7 +17,7 @@
 
 namespace xgboost {
 namespace tree {
-// Following 2 functions are slightly modifed version of fbcuda.
+// Following 2 functions are slightly modified version of fbcuda.
 
 /* \brief Constructs a rounding factor used to truncate elements in a sum such that the
    sum of the truncated elements is the same no matter what the order of the sum is.

src/tree/param.h

@@ -76,7 +76,7 @@ struct TrainParam : public XGBoostParameter<TrainParam> {
   // the criteria to use for ranking splits
   std::string split_evaluator;
 
-  // ------ From cpu quantile histogram -------.
+  // ------ From CPU quantile histogram -------.
   // percentage threshold for treating a feature as sparse
   // e.g. 0.2 indicates a feature with fewer than 20% nonzeros is considered sparse
   double sparse_threshold;
@@ -316,7 +316,7 @@ XGBOOST_DEVICE inline T CalcGain(const TrainingParams &p, StatT stat) {
   return CalcGain(p, stat.GetGrad(), stat.GetHess());
 }
 
-// Used in gpu code where GradientPair is used for gradient sum, not GradStats.
+// Used in GPU code where GradientPair is used for gradient sum, not GradStats.
 template <typename TrainingParams, typename GpairT>
 XGBOOST_DEVICE inline float CalcWeight(const TrainingParams &p, GpairT sum_grad) {
   return CalcWeight(p, sum_grad.GetGrad(), sum_grad.GetHess());
@@ -484,7 +484,7 @@ using SplitEntry = SplitEntryContainer<GradStats>;
 
 /*
  * \brief Parse the interaction constraints from string.
- * \param constraint_str String storing the interfaction constraints:
+ * \param constraint_str String storing the interaction constraints:
  *
  *  Example input string:
  *

src/tree/tree_model.cc

@@ -157,7 +157,7 @@ TreeGenerator* TreeGenerator::Create(std::string const& attrs, FeatureMap const&
   if (pos != std::string::npos) {
     name = attrs.substr(0, pos);
     params = attrs.substr(pos+1, attrs.length() - pos - 1);
-    // Eliminate all occurances of single quote string.
+    // Eliminate all occurrences of single quote string.
     size_t pos = std::string::npos;
     while ((pos = params.find('\'')) != std::string::npos) {
       params.replace(pos, 1, "\"");
@@ -1069,7 +1069,7 @@ void RegTree::CalculateContributionsApprox(const RegTree::FVec &feat,
 // Used by TreeShap
 // data we keep about our decision path
 // note that pweight is included for convenience and is not tied with the other attributes
-// the pweight of the i'th path element is the permuation weight of paths with i-1 ones in them
+// the pweight of the i'th path element is the permutation weight of paths with i-1 ones in them
 struct PathElement {
   int feature_index;
   bst_float zero_fraction;
@@ -1123,7 +1123,7 @@ void UnwindPath(PathElement *unique_path, unsigned unique_depth,
   }
 }
 
-// determine what the total permuation weight would be if
+// determine what the total permutation weight would be if
 // we unwound a previous extension in the decision path
 bst_float UnwoundPathSum(const PathElement *unique_path, unsigned unique_depth,
                          unsigned path_index) {

src/tree/updater_basemaker-inl.h

@@ -196,8 +196,8 @@ class BaseMaker: public TreeUpdater {
     }
   }
   /*!
-   * \brief this is helper function uses column based data structure,
-   *        reset the positions to the lastest one
+   * \brief This is a helper function that uses a column based data structure
+   *        and reset the positions to the latest one
    * \param nodes the set of nodes that contains the split to be used
    * \param p_fmat feature matrix needed for tree construction
    * \param tree the regression tree structure

src/tree/updater_gpu_hist.cu

@@ -549,7 +549,7 @@ struct GPUHistMakerDevice {
       bst_float weight = evaluator.CalcWeight(
           pos, param_d, GradStats{d_node_sum_gradients[pos]});
       static_assert(!std::is_const<decltype(out_preds_d)>::value, "");
-      auto v_predt = out_preds_d;  // for some reaon out_preds_d is const by both nvcc and clang.
+      auto v_predt = out_preds_d;  // for some reason out_preds_d is const by both nvcc and clang.
       v_predt[d_ridx[local_idx]] += weight * param_d.learning_rate;
     });
     row_partitioner.reset();

src/tree/updater_histmaker.cc

@@ -401,7 +401,7 @@ class CQHistMaker: public HistMaker {
     for (auto& sketch : sketchs_) {
       sketch.Init(info.num_row_, this->param_.sketch_eps);
     }
-    // intitialize the summary array
+    // initialize the summary array
     summary_array_.resize(sketchs_.size());
     // setup maximum size
     unsigned max_size = this->param_.MaxSketchSize();
@@ -409,7 +409,7 @@ class CQHistMaker: public HistMaker {
       summary_array_[i].Reserve(max_size);
     }
     {
-      // get smmary
+      // get summary
       thread_sketch_.resize(omp_get_max_threads());
 
       // TWOPASS: use the real set + split set in the column iteration.

src/tree/updater_quantile_hist.h

@@ -441,7 +441,7 @@ class QuantileHistMaker: public TreeUpdater {
     std::unique_ptr<ExpandQueue> qexpand_loss_guided_;
     std::vector<ExpandEntry> qexpand_depth_wise_;
     // key is the node id which should be calculated by Subtraction Trick, value is the node which
-    // provides the evidence for substracts
+    // provides the evidence for subtraction
     std::vector<ExpandEntry> nodes_for_subtraction_trick_;
     // list of nodes whose histograms would be built explicitly.
     std::vector<ExpandEntry> nodes_for_explicit_hist_build_;

src/tree/updater_refresh.cc

@@ -123,7 +123,7 @@ class TreeRefresher: public TreeUpdater {
     // start from groups that belongs to current data
     auto pid = 0;
     gstats[pid].Add(gpair[ridx]);
-    // tranverse tree
+    // traverse tree
     while (!tree[pid].IsLeaf()) {
       unsigned split_index = tree[pid].SplitIndex();
       pid = tree.GetNext(pid, feat.GetFvalue(split_index), feat.IsMissing(split_index));