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

lint half way

Browse Source
This commit is contained in:
tqchen 2015-07-03 18:31:52 -07:00
parent 2ed40523ab
commit 0162bb7034
21 changed files with 573 additions and 391 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
src/gbm/gblinear-inl.hpp
View File

@ -1,11 +1,13 @@
#ifndef XGBOOST_GBM_GBLINEAR_INL_HPP_
#define XGBOOST_GBM_GBLINEAR_INL_HPP_
/*!
* Copyright by Contributors
* \file gblinear-inl.hpp
* \brief Implementation of Linear booster, with L1/L2 regularization: Elastic Net
* the update rule is parallel coordinate descent (shotgun)
* \author Tianqi Chen
*/
#ifndef XGBOOST_GBM_GBLINEAR_INL_HPP_
#define XGBOOST_GBM_GBLINEAR_INL_HPP_
#include <vector>
#include <string>
#include <sstream>
@ -33,10 +35,10 @@ class GBLinear : public IGradBooster {
model.param.SetParam(name, val);
}
}
virtual void LoadModel(utils::IStream &fi, bool with_pbuffer) {
virtual void LoadModel(utils::IStream &fi, bool with_pbuffer) { // NOLINT(*)
model.LoadModel(fi);
}
virtual void SaveModel(utils::IStream &fo, bool with_pbuffer) const {
virtual void SaveModel(utils::IStream &fo, bool with_pbuffer) const { // NOLINT(*)
model.SaveModel(fo);
}
virtual void InitModel(void) {
@ -92,7 +94,8 @@ class GBLinear : public IGradBooster {
sum_hess += p.hess * v * v;
}
float &w = model[fid][gid];
bst_float dw = static_cast<bst_float>(param.learning_rate * param.CalcDelta(sum_grad, sum_hess, w));
bst_float dw = static_cast<bst_float>(param.learning_rate *
param.CalcDelta(sum_grad, sum_hess, w));
w += dw;
// update grad value
for (bst_uint j = 0; j < col.length; ++j) {
@ -258,12 +261,12 @@ class GBLinear : public IGradBooster {
std::fill(weight.begin(), weight.end(), 0.0f);
}
// save the model to file
inline void SaveModel(utils::IStream &fo) const {
inline void SaveModel(utils::IStream &fo) const { // NOLINT(*)
fo.Write(&param, sizeof(Param));
fo.Write(weight);
}
// load model from file
inline void LoadModel(utils::IStream &fi) {
inline void LoadModel(utils::IStream &fi) { // NOLINT(*)
utils::Assert(fi.Read(&param, sizeof(Param)) != 0, "Load LinearBooster");
fi.Read(&weight);
}

1
src/gbm/gbm.cpp
View File

@ -1,3 +1,4 @@
// Copyright by Contributors
#define _CRT_SECURE_NO_WARNINGS
#define _CRT_SECURE_NO_DEPRECATE
#define NOMINMAX

27
src/gbm/gbm.h
View File

@ -1,11 +1,14 @@
#ifndef XGBOOST_GBM_GBM_H_
#define XGBOOST_GBM_GBM_H_
/*!
* Copyright by Contributors
* \file gbm.h
* \brief interface of gradient booster, that learns through gradient statistics
* \author Tianqi Chen
*/
#ifndef XGBOOST_GBM_GBM_H_
#define XGBOOST_GBM_GBM_H_
#include <vector>
#include <string>
#include "../data.h"
#include "../utils/io.h"
#include "../utils/fmap.h"
@ -13,7 +16,7 @@
namespace xgboost {
/*! \brief namespace for gradient booster */
namespace gbm {
/*!
/*!
* \brief interface of gradient boosting model
*/
class IGradBooster {
@ -29,26 +32,26 @@ class IGradBooster {
* \param fi input stream
* \param with_pbuffer whether the incoming data contains pbuffer
*/
virtual void LoadModel(utils::IStream &fi, bool with_pbuffer) = 0;
virtual void LoadModel(utils::IStream &fi, bool with_pbuffer) = 0; // NOLINT(*)
/*!
* \brief save model to stream
* \param fo output stream
* \param with_pbuffer whether save out pbuffer
*/
virtual void SaveModel(utils::IStream &fo, bool with_pbuffer) const = 0;
virtual void SaveModel(utils::IStream &fo, bool with_pbuffer) const = 0; // NOLINT(*)
/*!
* \brief initialize the model
*/
virtual void InitModel(void) = 0;
/*!
/*!
* \brief reset the predict buffer
* this will invalidate all the previous cached results
* and recalculate from scratch
*/
virtual void ResetPredBuffer(size_t num_pbuffer) {}
/*!
/*!
* \brief whether the model allow lazy checkpoint
* return true if model is only updated in DoBoost
* return true if model is only updated in DoBoost
* after all Allreduce calls
*/
virtual bool AllowLazyCheckPoint(void) const {
@ -76,20 +79,20 @@ class IGradBooster {
* the size of buffer is set by convention using IGradBooster.SetParam("num_pbuffer","size")
* \param info extra side information that may be needed for prediction
* \param out_preds output vector to hold the predictions
* \param ntree_limit limit the number of trees used in prediction, when it equals 0, this means
* \param ntree_limit limit the number of trees used in prediction, when it equals 0, this means
* we do not limit number of trees, this parameter is only valid for gbtree, but not for gblinear
*/
virtual void Predict(IFMatrix *p_fmat,
int64_t buffer_offset,
const BoosterInfo &info,
std::vector<float> *out_preds,
unsigned ntree_limit = 0) = 0;
unsigned ntree_limit = 0) = 0;
/*!
* \brief online prediction funciton, predict score for one instance at a time
* NOTE: use the batch prediction interface if possible, batch prediction is usually
* more efficient than online prediction
* This function is NOT threadsafe, make sure you only call from one thread
*
*
* \param inst the instance you want to predict
* \param out_preds output vector to hold the predictions
* \param ntree_limit limit the number of trees used in prediction
@ -106,7 +109,7 @@ class IGradBooster {
* \param p_fmat feature matrix
* \param info extra side information that may be needed for prediction
* \param out_preds output vector to hold the predictions
* \param ntree_limit limit the number of trees used in prediction, when it equals 0, this means
* \param ntree_limit limit the number of trees used in prediction, when it equals 0, this means
* we do not limit number of trees, this parameter is only valid for gbtree, but not for gblinear
*/
virtual void PredictLeaf(IFMatrix *p_fmat,

46
src/gbm/gbtree-inl.hpp
View File

@ -1,13 +1,16 @@
#ifndef XGBOOST_GBM_GBTREE_INL_HPP_
#define XGBOOST_GBM_GBTREE_INL_HPP_
/*!
* Copyright by Contributors
* \file gbtree-inl.hpp
* \brief gradient boosted tree implementation
* \author Tianqi Chen
*/
#ifndef XGBOOST_GBM_GBTREE_INL_HPP_
#define XGBOOST_GBM_GBTREE_INL_HPP_
#include <vector>
#include <utility>
#include <string>
#include <limits>
#include "./gbm.h"
#include "../utils/omp.h"
#include "../tree/updater.h"
@ -39,7 +42,7 @@ class GBTree : public IGradBooster {
tparam.SetParam(name, val);
if (trees.size() == 0) mparam.SetParam(name, val);
}
virtual void LoadModel(utils::IStream &fi, bool with_pbuffer) {
virtual void LoadModel(utils::IStream &fi, bool with_pbuffer) { // NOLINT(*)
this->Clear();
utils::Check(fi.Read(&mparam, sizeof(ModelParam)) != 0,
"GBTree: invalid model file");
@ -62,10 +65,10 @@ class GBTree : public IGradBooster {
"GBTree: invalid model file");
}
}
virtual void SaveModel(utils::IStream &fo, bool with_pbuffer) const {
virtual void SaveModel(utils::IStream &fo, bool with_pbuffer) const { // NOLINT(*)
utils::Assert(mparam.num_trees == static_cast<int>(trees.size()), "GBTree");
if (with_pbuffer) {
fo.Write(&mparam, sizeof(ModelParam));
fo.Write(&mparam, sizeof(ModelParam));
} else {
ModelParam p = mparam;
p.num_pbuffer = 0;
@ -129,7 +132,7 @@ class GBTree : public IGradBooster {
int64_t buffer_offset,
const BoosterInfo &info,
std::vector<float> *out_preds,
unsigned ntree_limit = 0) {
unsigned ntree_limit = 0) {
int nthread;
#pragma omp parallel
{
@ -160,12 +163,12 @@ class GBTree : public IGradBooster {
this->Pred(batch[i],
buffer_offset < 0 ? -1 : buffer_offset + ridx,
gid, info.GetRoot(ridx), &feats,
&preds[ridx * mparam.num_output_group + gid], stride,
&preds[ridx * mparam.num_output_group + gid], stride,
ntree_limit);
}
}
}
}
}
virtual void Predict(const SparseBatch::Inst &inst,
std::vector<float> *out_preds,
unsigned ntree_limit,
@ -178,10 +181,10 @@ class GBTree : public IGradBooster {
// loop over output groups
for (int gid = 0; gid < mparam.num_output_group; ++gid) {
this->Pred(inst, -1, gid, root_index, &thread_temp[0],
&(*out_preds)[gid], mparam.num_output_group,
&(*out_preds)[gid], mparam.num_output_group,
ntree_limit);
}
}
}
virtual void PredictLeaf(IFMatrix *p_fmat,
const BoosterInfo &info,
std::vector<float> *out_preds,
@ -196,7 +199,6 @@ class GBTree : public IGradBooster {
thread_temp[i].Init(mparam.num_feature);
}
this->PredPath(p_fmat, info, out_preds, ntree_limit);
}
virtual std::vector<std::string> DumpModel(const utils::FeatMap& fmap, int option) {
std::vector<std::string> dump;
@ -260,7 +262,7 @@ class GBTree : public IGradBooster {
// update the trees
for (size_t i = 0; i < updaters.size(); ++i) {
updaters[i]->Update(gpair, p_fmat, info, new_trees);
}
}
// optimization, update buffer, if possible
// this is only under distributed column mode
// for safety check of lazy checkpoint
@ -287,7 +289,7 @@ class GBTree : public IGradBooster {
}
// update buffer by pre-cached position
inline void UpdateBufferByPosition(IFMatrix *p_fmat,
int64_t buffer_offset,
int64_t buffer_offset,
int bst_group,
const tree::RegTree &new_tree,
const int* leaf_position) {
@ -313,11 +315,11 @@ class GBTree : public IGradBooster {
int bst_group,
unsigned root_index,
tree::RegTree::FVec *p_feats,
float *out_pred, size_t stride,
float *out_pred, size_t stride,
unsigned ntree_limit) {
size_t itop = 0;
float psum = 0.0f;
// sum of leaf vector
// sum of leaf vector
std::vector<float> vec_psum(mparam.size_leaf_vector, 0.0f);
const int64_t bid = mparam.BufferOffset(buffer_index, bst_group);
// number of valid trees
@ -339,7 +341,7 @@ class GBTree : public IGradBooster {
for (int j = 0; j < mparam.size_leaf_vector; ++j) {
vec_psum[j] += trees[i]->leafvec(tid)[j];
}
if(--treeleft == 0) break;
if (--treeleft == 0) break;
}
}
p_feats->Drop(inst);
@ -365,7 +367,7 @@ class GBTree : public IGradBooster {
// number of valid trees
if (ntree_limit == 0 || ntree_limit > trees.size()) {
ntree_limit = static_cast<unsigned>(trees.size());
}
}
std::vector<float> &preds = *out_preds;
preds.resize(info.num_row * ntree_limit);
// start collecting the prediction
@ -389,7 +391,7 @@ class GBTree : public IGradBooster {
}
}
}
// --- data structure ---
/*! \brief training parameters */
struct TrainParam {
@ -442,10 +444,10 @@ class GBTree : public IGradBooster {
int num_feature;
/*! \brief size of predicton buffer allocated used for buffering */
int64_t num_pbuffer;
/*!
/*!
* \brief how many output group a single instance can produce
* this affects the behavior of number of output we have:
* suppose we have n instance and k group, output will be k*n
* suppose we have n instance and k group, output will be k*n
*/
int num_output_group;
/*! \brief size of leaf vector needed in tree */
@ -478,8 +480,8 @@ class GBTree : public IGradBooster {
inline size_t PredBufferSize(void) const {
return num_output_group * num_pbuffer * (size_leaf_vector + 1);
}
/*!
* \brief get the buffer offset given a buffer index and group id
/*!
* \brief get the buffer offset given a buffer index and group id
* \return calculated buffer offset
*/
inline int64_t BufferOffset(int64_t buffer_index, int bst_group) const {

9
src/io/io.h
View File

@ -1,11 +1,13 @@
#ifndef XGBOOST_IO_IO_H_
#define XGBOOST_IO_IO_H_
/*!
* Copyright 2014 by Contributors
* \file io.h
* \brief handles input data format of xgboost
* I/O module handles a specific DMatrix format
* \author Tianqi Chen
*/
#ifndef XGBOOST_IO_IO_H_
#define XGBOOST_IO_IO_H_
#include "../data.h"
#include "../learner/dmatrix.h"
@ -32,7 +34,7 @@ DataMatrix* LoadDataMatrix(const char *fname,
bool loadsplit,
const char *cache_file = NULL);
/*!
* \brief save DataMatrix into stream,
* \brief save DataMatrix into stream,
* note: the saved dmatrix format may not be in exactly same as input
* SaveDMatrix will choose the best way to materialize the dmatrix.
* \param dmat the dmatrix to be saved
@ -40,7 +42,6 @@ DataMatrix* LoadDataMatrix(const char *fname,
* \param silent whether print message during saving
*/
void SaveDataMatrix(const DataMatrix &dmat, const char *fname, bool silent = false);
} // namespace io
} // namespace xgboost
#endif // XGBOOST_IO_IO_H_

80
src/tree/model.h
View File

@ -1,10 +1,12 @@
#ifndef XGBOOST_TREE_MODEL_H_
#define XGBOOST_TREE_MODEL_H_
/*!
* Copyright 2014 by Contributors
* \file model.h
* \brief model structure for tree
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_MODEL_H_
#define XGBOOST_TREE_MODEL_H_
#include <string>
#include <cstring>
#include <sstream>
@ -19,7 +21,7 @@
namespace xgboost {
namespace tree {
/*!
* \brief template class of TreeModel
* \brief template class of TreeModel
* \tparam TSplitCond data type to indicate split condition
* \tparam TNodeStat auxiliary statistics of node to help tree building
*/
@ -42,7 +44,7 @@ class TreeModel {
int max_depth;
/*! \brief number of features used for tree construction */
int num_feature;
/*!
/*!
* \brief leaf vector size, used for vector tree
* used to store more than one dimensional information in tree
*/
@ -55,8 +57,8 @@ class TreeModel {
size_leaf_vector = 0;
std::memset(reserved, 0, sizeof(reserved));
}
/*!
* \brief set parameters from outside
/*!
* \brief set parameters from outside
* \param name name of the parameter
* \param val value of the parameter
*/
@ -70,7 +72,7 @@ class TreeModel {
/*! \brief tree node */
class Node {
public:
Node(void) : sindex_(0) {}
Node(void) : sindex_(0) {}
/*! \brief index of left child */
inline int cleft(void) const {
return this->cleft_;
@ -119,15 +121,15 @@ class TreeModel {
inline bool is_root(void) const {
return parent_ == -1;
}
/*!
* \brief set the right child
/*!
* \brief set the right child
* \param nide node id to right child
*/
inline void set_right_child(int nid) {
this->cright_ = nid;
}
/*!
* \brief set split condition of current node
/*!
* \brief set split condition of current node
* \param split_index feature index to split
* \param split_cond split condition
* \param default_left the default direction when feature is unknown
@ -138,10 +140,10 @@ class TreeModel {
this->sindex_ = split_index;
(this->info_).split_cond = split_cond;
}
/*!
/*!
* \brief set the leaf value of the node
* \param value leaf value
* \param right right index, could be used to store
* \param right right index, could be used to store
* additional information
*/
inline void set_leaf(float value, int right = -1) {
@ -153,12 +155,12 @@ class TreeModel {
inline void mark_delete(void) {
this->sindex_ = std::numeric_limits<unsigned>::max();
}
private:
friend class TreeModel<TSplitCond, TNodeStat>;
/*!
* \brief in leaf node, we have weights, in non-leaf nodes,
* we have split condition
/*!
* \brief in leaf node, we have weights, in non-leaf nodes,
* we have split condition
*/
union Info{
float leaf_value;
@ -203,7 +205,7 @@ class TreeModel {
"number of nodes in the tree exceed 2^31");
nodes.resize(param.num_nodes);
stats.resize(param.num_nodes);
leaf_vector.resize(param.num_nodes * param.size_leaf_vector);
leaf_vector.resize(param.num_nodes * param.size_leaf_vector);
return nd;
}
// delete a tree node, keep the parent field to allow trace back
@ -215,7 +217,7 @@ class TreeModel {
}
public:
/*!
/*!
* \brief change a non leaf node to a leaf node, delete its children
* \param rid node id of the node
* \param new leaf value
@ -229,7 +231,7 @@ class TreeModel {
this->DeleteNode(nodes[rid].cright());
nodes[rid].set_leaf(value);
}
/*!
/*!
* \brief collapse a non leaf node to a leaf node, delete its children
* \param rid node id of the node
* \param new leaf value
@ -273,7 +275,7 @@ class TreeModel {
return &leaf_vector[nid * param.size_leaf_vector];
}
/*! \brief get leaf vector given nid */
inline const bst_float* leafvec(int nid) const{
inline const bst_float* leafvec(int nid) const {
if (leaf_vector.size() == 0) return NULL;
return &leaf_vector[nid * param.size_leaf_vector];
}
@ -288,15 +290,15 @@ class TreeModel {
nodes[i].set_parent(-1);
}
}
/*!
/*!
* \brief load model from stream
* \param fi input stream
*/
inline void LoadModel(utils::IStream &fi) {
inline void LoadModel(utils::IStream &fi) { // NOLINT(*)
utils::Check(fi.Read(&param, sizeof(Param)) > 0,
"TreeModel: wrong format");
nodes.resize(param.num_nodes); stats.resize(param.num_nodes);
utils::Assert(param.num_nodes != 0, "invalid model");
utils::Assert(param.num_nodes != 0, "invalid model");
utils::Check(fi.Read(BeginPtr(nodes), sizeof(Node) * nodes.size()) > 0,
"TreeModel: wrong format");
utils::Check(fi.Read(BeginPtr(stats), sizeof(NodeStat) * stats.size()) > 0,
@ -313,22 +315,22 @@ class TreeModel {
"number of deleted nodes do not match, num_deleted=%d, dnsize=%lu, num_nodes=%d",
param.num_deleted, deleted_nodes.size(), param.num_nodes);
}
/*!
/*!
* \brief save model to stream
* \param fo output stream
*/
inline void SaveModel(utils::IStream &fo) const {
inline void SaveModel(utils::IStream &fo) const { // NOLINT(*)
utils::Assert(param.num_nodes == static_cast<int>(nodes.size()),
"Tree::SaveModel");
utils::Assert(param.num_nodes == static_cast<int>(stats.size()),
"Tree::SaveModel");
fo.Write(&param, sizeof(Param));
utils::Assert(param.num_nodes != 0, "invalid model");
utils::Assert(param.num_nodes != 0, "invalid model");
fo.Write(BeginPtr(nodes), sizeof(Node) * nodes.size());
fo.Write(BeginPtr(stats), sizeof(NodeStat) * nodes.size());
if (param.size_leaf_vector != 0) fo.Write(leaf_vector);
}
/*!
/*!
* \brief add child nodes to node
* \param nid node id to add childs
*/
@ -340,8 +342,8 @@ class TreeModel {
nodes[nodes[nid].cleft() ].set_parent(nid, true);
nodes[nodes[nid].cright()].set_parent(nid, false);
}
/*!
* \brief only add a right child to a leaf node
/*!
* \brief only add a right child to a leaf node
* \param node id to add right child
*/
inline void AddRightChild(int nid) {
@ -385,7 +387,7 @@ class TreeModel {
inline int num_extra_nodes(void) const {
return param.num_nodes - param.num_roots - param.num_deleted;
}
/*!
/*!
* \brief dump model to text string
* \param fmap feature map of feature types
* \param with_stats whether dump out statistics as well
@ -400,7 +402,7 @@ class TreeModel {
}
private:
void Dump(int nid, std::stringstream &fo,
void Dump(int nid, std::stringstream &fo, // NOLINT(*)
const utils::FeatMap& fmap, int depth, bool with_stats) {
for (int i = 0; i < depth; ++i) {
fo << '\t';
@ -469,7 +471,7 @@ struct RTreeNodeStat {
/*! \brief number of child that is leaf node known up to now */
int leaf_child_cnt;
/*! \brief print information of current stats to fo */
inline void Print(std::stringstream &fo, bool is_leaf) const {
inline void Print(std::stringstream &fo, bool is_leaf) const { // NOLINT(*)
if (!is_leaf) {
fo << ",gain=" << loss_chg << ",cover=" << sum_hess;
} else {
@ -481,13 +483,13 @@ struct RTreeNodeStat {
/*! \brief define regression tree to be the most common tree model */
class RegTree: public TreeModel<bst_float, RTreeNodeStat>{
public:
/*!
/*!
* \brief dense feature vector that can be taken by RegTree
* to do tranverse efficiently
* and can be construct from sparse feature vector
*/
struct FVec {
/*!
/*!
* \brief a union value of value and flag
* when flag == -1, this indicate the value is missing
*/
@ -510,7 +512,7 @@ class RegTree: public TreeModel<bst_float, RTreeNodeStat>{
}
}
/*! \brief drop the trace after fill, must be called after fill */
inline void Drop(const RowBatch::Inst &inst) {
inline void Drop(const RowBatch::Inst &inst) {
for (bst_uint i = 0; i < inst.length; ++i) {
if (inst[i].index >= data.size()) continue;
data[inst[i].index].flag = -1;
@ -526,10 +528,10 @@ class RegTree: public TreeModel<bst_float, RTreeNodeStat>{
}
};
/*!
* \brief get the leaf index
* \brief get the leaf index
* \param feats dense feature vector, if the feature is missing the field is set to NaN
* \param root_gid starting root index of the instance
* \return the leaf index of the given feature
* \return the leaf index of the given feature
*/
inline int GetLeafIndex(const FVec&feat, unsigned root_id = 0) const {
// start from groups that belongs to current data
@ -545,7 +547,7 @@ class RegTree: public TreeModel<bst_float, RTreeNodeStat>{
* \brief get the prediction of regression tree, only accepts dense feature vector
* \param feats dense feature vector, if the feature is missing the field is set to NaN
* \param root_gid starting root index of the instance
* \return the leaf index of the given feature
* \return the leaf index of the given feature
*/
inline float Predict(const FVec &feat, unsigned root_id = 0) const {
int pid = this->GetLeafIndex(feat, root_id);

49
src/tree/param.h
View File

@ -1,10 +1,13 @@
#ifndef XGBOOST_TREE_PARAM_H_
#define XGBOOST_TREE_PARAM_H_
/*!
* Copyright 2014 by Contributors
* \file param.h
* \brief training parameters, statistics used to support tree construction
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_PARAM_H_
#define XGBOOST_TREE_PARAM_H_
#include <vector>
#include <cstring>
#include "../data.h"
@ -27,7 +30,7 @@ struct TrainParam{
// L1 regularization factor
float reg_alpha;
// default direction choice
int default_direction;
int default_direction;
// maximum delta update we can add in weight estimation
// this parameter can be used to stablize update
// default=0 means no constraint on weight delta
@ -45,7 +48,7 @@ struct TrainParam{
// accuracy of sketch
float sketch_ratio;
// leaf vector size
int size_leaf_vector;
int size_leaf_vector;
// option for parallelization
int parallel_option;
// option to open cacheline optimizaton
@ -74,11 +77,11 @@ struct TrainParam{
sketch_ratio = 2.0f;
cache_opt = 1;
}
/*!
* \brief set parameters from outside
/*!
* \brief set parameters from outside
* \param name name of the parameter
* \param val value of the parameter
*/
*/
inline void SetParam(const char *name, const char *val) {
using namespace std;
// sync-names
@ -116,7 +119,7 @@ struct TrainParam{
if (reg_alpha == 0.0f) {
return Sqr(sum_grad) / (sum_hess + reg_lambda);
} else {
return Sqr(ThresholdL1(sum_grad, reg_alpha)) / (sum_hess + reg_lambda);
return Sqr(ThresholdL1(sum_grad, reg_alpha)) / (sum_hess + reg_lambda);
}
} else {
double w = CalcWeight(sum_grad, sum_hess);
@ -213,7 +216,7 @@ struct GradStats {
inline static void CheckInfo(const BoosterInfo &info) {
}
/*!
* \brief accumulate statistics
* \brief accumulate statistics
* \param p the gradient pair
*/
inline void Add(bst_gpair p) {
@ -222,7 +225,7 @@ struct GradStats {
/*!
* \brief accumulate statistics, more complicated version
* \param gpair the vector storing the gradient statistics
* \param info the additional information
* \param info the additional information
* \param ridx instance index of this instance
*/
inline void Add(const std::vector<bst_gpair> &gpair,
@ -244,7 +247,7 @@ struct GradStats {
this->Add(b.sum_grad, b.sum_hess);
}
/*! \brief same as add, reduce is used in All Reduce */
inline static void Reduce(GradStats &a, const GradStats &b) {
inline static void Reduce(GradStats &a, const GradStats &b) { // NOLINT(*)
a.Add(b);
}
/*! \brief set current value to a - b */
@ -257,8 +260,8 @@ struct GradStats {
return sum_hess == 0.0;
}
/*! \brief set leaf vector value based on statistics */
inline void SetLeafVec(const TrainParam &param, bst_float *vec) const{
}
inline void SetLeafVec(const TrainParam &param, bst_float *vec) const {
}
// constructor to allow inheritance
GradStats(void) {}
/*! \brief add statistics to the data */
@ -311,7 +314,7 @@ struct CVGradStats : public GradStats {
ret += param.CalcGain(train[i].sum_grad,
train[i].sum_hess,
vsize * valid[i].sum_grad,
vsize * valid[i].sum_hess);
vsize * valid[i].sum_hess);
}
return ret / vsize;
}
@ -324,7 +327,7 @@ struct CVGradStats : public GradStats {
}
}
/*! \brief same as add, reduce is used in All Reduce */
inline static void Reduce(CVGradStats &a, const CVGradStats &b) {
inline static void Reduce(CVGradStats &a, const CVGradStats &b) { // NOLINT(*)
a.Add(b);
}
/*! \brief set current value to a - b */
@ -344,8 +347,8 @@ struct CVGradStats : public GradStats {
}
};
/*!
* \brief statistics that is helpful to store
/*!
* \brief statistics that is helpful to store
* and represent a split solution for the tree
*/
struct SplitEntry{
@ -357,12 +360,12 @@ struct SplitEntry{
float split_value;
/*! \brief constructor */
SplitEntry(void) : loss_chg(0.0f), sindex(0), split_value(0.0f) {}
/*!
* \brief decides whether a we can replace current entry with the statistics given
/*!
* \brief decides whether a we can replace current entry with the statistics given
* This function gives better priority to lower index when loss_chg equals
* not the best way, but helps to give consistent result during multi-thread execution
* \param loss_chg the loss reduction get through the split
* \param split_index the feature index where the split is on
* \param split_index the feature index where the split is on
*/
inline bool NeedReplace(bst_float new_loss_chg, unsigned split_index) const {
if (this->split_index() <= split_index) {
@ -371,7 +374,7 @@ struct SplitEntry{
return !(this->loss_chg > new_loss_chg);
}
}
/*!
/*!
* \brief update the split entry, replace it if e is better
* \param e candidate split solution
* \return whether the proposed split is better and can replace current split
@ -386,7 +389,7 @@ struct SplitEntry{
return false;
}
}
/*!
/*!
* \brief update the split entry, replace it if e is better
* \param loss_chg loss reduction of new candidate
* \param split_index feature index to split on
@ -407,7 +410,7 @@ struct SplitEntry{
}
}
/*! \brief same as update, used by AllReduce*/
inline static void Reduce(SplitEntry &dst, const SplitEntry &src) {
inline static void Reduce(SplitEntry &dst, const SplitEntry &src) { // NOLINT(*)
dst.Update(src);
}
/*!\return feature index to split on */

1
src/tree/updater.cpp
View File

@ -1,3 +1,4 @@
// Copyright 2014 by Contributors
#define _CRT_SECURE_NO_WARNINGS
#define _CRT_SECURE_NO_DEPRECATE
#define NOMINMAX

20
src/tree/updater.h
View File

@ -1,10 +1,12 @@
#ifndef XGBOOST_TREE_UPDATER_H_
#define XGBOOST_TREE_UPDATER_H_
/*!
* Copyright 2014 by Contributors
* \file updater.h
* \brief interface to update the tree
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_UPDATER_H_
#define XGBOOST_TREE_UPDATER_H_
#include <vector>
#include "../data.h"
@ -12,7 +14,7 @@
namespace xgboost {
namespace tree {
/*!
/*!
* \brief interface of tree update module, that performs update of a tree
*/
class IUpdater {
@ -21,7 +23,7 @@ class IUpdater {
* \brief set parameters from outside
* \param name name of the parameter
* \param val value of the parameter
*/
*/
virtual void SetParam(const char *name, const char *val) = 0;
/*!
* \brief peform update to the tree models
@ -29,8 +31,8 @@ class IUpdater {
* \param p_fmat feature matrix that provide access to features
* \param info extra side information that may be need, such as root index
* \param trees pointer to the trees to be updated, upater will change the content of the tree
* note: all the trees in the vector are updated, with the same statistics,
* but maybe different random seeds, usually one tree is passed in at a time,
* note: all the trees in the vector are updated, with the same statistics,
* but maybe different random seeds, usually one tree is passed in at a time,
* there can be multiple trees when we train random forest style model
*/
virtual void Update(const std::vector<bst_gpair> &gpair,
@ -38,7 +40,7 @@ class IUpdater {
const BoosterInfo &info,
const std::vector<RegTree*> &trees) = 0;
/*!
/*!
* \brief this is simply a function for optimizing performance
* this function asks the updater to return the leaf position of each instance in the p_fmat,
* if it is cached in the updater, if it is not available, return NULL
@ -50,8 +52,8 @@ class IUpdater {
// destructor
virtual ~IUpdater(void) {}
};
/*!
* \brief create a updater based on name
/*!
* \brief create a updater based on name
* \param name name of updater
* \return return the updater instance
*/

77
src/tree/updater_basemaker-inl.hpp
View File

@ -1,12 +1,14 @@
#ifndef XGBOOST_TREE_UPDATER_BASEMAKER_INL_HPP_
#define XGBOOST_TREE_UPDATER_BASEMAKER_INL_HPP_
/*!
* Copyright 2014 by Contributors
* \file updater_basemaker-inl.hpp
* \brief implement a common tree constructor
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_UPDATER_BASEMAKER_INL_HPP_
#define XGBOOST_TREE_UPDATER_BASEMAKER_INL_HPP_
#include <vector>
#include <algorithm>
#include <string>
#include <limits>
#include "../sync/sync.h"
#include "../utils/random.h"
@ -14,7 +16,7 @@
namespace xgboost {
namespace tree {
/*!
/*!
* \brief base tree maker class that defines common operation
* needed in tree making
*/
@ -26,7 +28,7 @@ class BaseMaker: public IUpdater {
virtual void SetParam(const char *name, const char *val) {
param.SetParam(name, val);
}
protected:
// helper to collect and query feature meta information
struct FMetaHelper {
@ -60,8 +62,11 @@ class BaseMaker: public IUpdater {
bst_float a = fminmax[fid * 2];
bst_float b = fminmax[fid * 2 + 1];
if (a == -std::numeric_limits<bst_float>::max()) return 0;
if (-a == b) return 1;
else return 2;
if (-a == b) {
return 1;
} else {
return 2;
}
}
inline bst_float MaxValue(bst_uint fid) const {
return fminmax[fid *2 + 1];
@ -70,7 +75,7 @@ class BaseMaker: public IUpdater {
std::vector<bst_uint> &findex = *p_findex;
findex.clear();
for (size_t i = 0; i < fminmax.size(); i += 2) {
const bst_uint fid = static_cast<bst_uint>(i / 2);
const bst_uint fid = static_cast<bst_uint>(i / 2);
if (this->Type(fid) != 0) findex.push_back(fid);
}
unsigned n = static_cast<unsigned>(p * findex.size());
@ -86,7 +91,7 @@ class BaseMaker: public IUpdater {
rabit::Broadcast(&s_cache, 0);
fs.Read(&findex);
}
private:
std::vector<bst_float> fminmax;
};
@ -116,7 +121,7 @@ class BaseMaker: public IUpdater {
}
return nthread;
}
// ------class member helpers---------
// ------class member helpers---------
/*! \brief initialize temp data structure */
inline void InitData(const std::vector<bst_gpair> &gpair,
const IFMatrix &fmat,
@ -124,7 +129,8 @@ class BaseMaker: public IUpdater {
const RegTree &tree) {
utils::Assert(tree.param.num_nodes == tree.param.num_roots,
"TreeMaker: can only grow new tree");
{// setup position
{
// setup position
position.resize(gpair.size());
if (root_index.size() == 0) {
std::fill(position.begin(), position.end(), 0);
@ -147,7 +153,8 @@ class BaseMaker: public IUpdater {
}
}
}
{// expand query
{
// expand query
qexpand.reserve(256); qexpand.clear();
for (int i = 0; i < tree.param.num_roots; ++i) {
qexpand.push_back(i);
@ -170,7 +177,7 @@ class BaseMaker: public IUpdater {
this->UpdateNode2WorkIndex(tree);
}
// return decoded position
inline int DecodePosition(bst_uint ridx) const{
inline int DecodePosition(bst_uint ridx) const {
const int pid = position[ridx];
return pid < 0 ? ~pid : pid;
}
@ -182,23 +189,24 @@ class BaseMaker: public IUpdater {
position[ridx] = nid;
}
}
/*!
/*!
* \brief this is helper function uses column based data structure,
* reset the positions to the lastest 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
*/
inline void ResetPositionCol(const std::vector<int> &nodes, IFMatrix *p_fmat, const RegTree &tree) {
inline void ResetPositionCol(const std::vector<int> &nodes,
IFMatrix *p_fmat, const RegTree &tree) {
// set the positions in the nondefault
this->SetNonDefaultPositionCol(nodes, p_fmat, tree);
// set rest of instances to default position
const std::vector<bst_uint> &rowset = p_fmat->buffered_rowset();
// set default direct nodes to default
// for leaf nodes that are not fresh, mark then to ~nid,
// for leaf nodes that are not fresh, mark then to ~nid,
// so that they are ignored in future statistics collection
const bst_omp_uint ndata = static_cast<bst_omp_uint>(rowset.size());
#pragma omp parallel for schedule(static)
for (bst_omp_uint i = 0; i < ndata; ++i) {
const bst_uint ridx = rowset[i];
@ -237,7 +245,7 @@ class BaseMaker: public IUpdater {
}
std::sort(fsplits.begin(), fsplits.end());
fsplits.resize(std::unique(fsplits.begin(), fsplits.end()) - fsplits.begin());
utils::IIterator<ColBatch> *iter = p_fmat->ColIterator(fsplits);
while (iter->Next()) {
const ColBatch &batch = iter->Value();
@ -252,7 +260,7 @@ class BaseMaker: public IUpdater {
const int nid = this->DecodePosition(ridx);
// go back to parent, correct those who are not default
if (!tree[nid].is_leaf() && tree[nid].split_index() == fid) {
if(fvalue < tree[nid].split_cond()) {
if (fvalue < tree[nid].split_cond()) {
this->SetEncodePosition(ridx, tree[nid].cleft());
} else {
this->SetEncodePosition(ridx, tree[nid].cright());
@ -324,7 +332,7 @@ class BaseMaker: public IUpdater {
sketch->temp.size = 0;
}
/*!
* \brief push a new element to sketch
* \brief push a new element to sketch
* \param fvalue feature value, comes in sorted ascending order
* \param w weight
* \param max_size
@ -337,31 +345,32 @@ class BaseMaker: public IUpdater {
return;
}
if (last_fvalue != fvalue) {
double rmax = rmin + wmin;
double rmax = rmin + wmin;
if (rmax >= next_goal && sketch->temp.size != max_size) {
if (sketch->temp.size == 0 || last_fvalue > sketch->temp.data[sketch->temp.size-1].value) {
if (sketch->temp.size == 0 ||
last_fvalue > sketch->temp.data[sketch->temp.size-1].value) {
// push to sketch
sketch->temp.data[sketch->temp.size] =
utils::WXQuantileSketch<bst_float, bst_float>::
Entry(static_cast<bst_float>(rmin),
static_cast<bst_float>(rmax),
static_cast<bst_float>(wmin), last_fvalue);
static_cast<bst_float>(rmax),
static_cast<bst_float>(wmin), last_fvalue);
utils::Assert(sketch->temp.size < max_size,
"invalid maximum size max_size=%u, stemp.size=%lu\n",
max_size, sketch->temp.size);
++sketch->temp.size;
}
if (sketch->temp.size == max_size) {
next_goal = sum_total * 2.0f + 1e-5f;
} else{
next_goal = sum_total * 2.0f + 1e-5f;
} else {
next_goal = static_cast<bst_float>(sketch->temp.size * sum_total / max_size);
}
} else {
if (rmax >= next_goal) {
rabit::TrackerPrintf("INFO: rmax=%g, sum_total=%g, next_goal=%g, size=%lu\n",
rmax, sum_total, next_goal, sketch->temp.size);
}
}
if (rmax >= next_goal) {
rabit::TrackerPrintf("INFO: rmax=%g, sum_total=%g, next_goal=%g, size=%lu\n",
rmax, sum_total, next_goal, sketch->temp.size);
}
}
rmin = rmax;
wmin = w;
last_fvalue = fvalue;
@ -375,13 +384,13 @@ class BaseMaker: public IUpdater {
if (sketch->temp.size == 0 || last_fvalue > sketch->temp.data[sketch->temp.size-1].value) {
utils::Assert(sketch->temp.size <= max_size,
"Finalize: invalid maximum size, max_size=%u, stemp.size=%lu",
sketch->temp.size, max_size );
sketch->temp.size, max_size);
// push to sketch
sketch->temp.data[sketch->temp.size] =
utils::WXQuantileSketch<bst_float, bst_float>::
Entry(static_cast<bst_float>(rmin),
static_cast<bst_float>(rmax),
static_cast<bst_float>(wmin), last_fvalue);
static_cast<bst_float>(rmax),
static_cast<bst_float>(wmin), last_fvalue);
++sketch->temp.size;
}
sketch->PushTemp();
@ -415,4 +424,4 @@ class BaseMaker: public IUpdater {
};
} // namespace tree
} // namespace xgboost
#endif // XGBOOST_TREE_UPDATER_BASEMAKER_INL_HPP_
#endif // XGBOOST_TREE_UPDATER_BASEMAKER_INL_HPP_

141
src/tree/updater_colmaker-inl.hpp
View File

@ -1,10 +1,12 @@
#ifndef XGBOOST_TREE_UPDATER_COLMAKER_INL_HPP_
#define XGBOOST_TREE_UPDATER_COLMAKER_INL_HPP_
/*!
* Copyright 2014 by Contributors
* \file updater_colmaker-inl.hpp
* \brief use columnwise update to construct a tree
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_UPDATER_COLMAKER_INL_HPP_
#define XGBOOST_TREE_UPDATER_COLMAKER_INL_HPP_
#include <vector>
#include <cmath>
#include <algorithm>
@ -114,10 +116,13 @@ class ColMaker: public IUpdater {
// initialize temp data structure
inline void InitData(const std::vector<bst_gpair> &gpair,
const IFMatrix &fmat,
const std::vector<unsigned> &root_index, const RegTree &tree) {
utils::Assert(tree.param.num_nodes == tree.param.num_roots, "ColMaker: can only grow new tree");
const std::vector<unsigned> &root_index,
const RegTree &tree) {
utils::Assert(tree.param.num_nodes == tree.param.num_roots,
"ColMaker: can only grow new tree");
const std::vector<bst_uint> &rowset = fmat.buffered_rowset();
{// setup position
{
// setup position
position.resize(gpair.size());
if (root_index.size() == 0) {
for (size_t i = 0; i < rowset.size(); ++i) {
@ -127,7 +132,8 @@ class ColMaker: public IUpdater {
for (size_t i = 0; i < rowset.size(); ++i) {
const bst_uint ridx = rowset[i];
position[ridx] = root_index[ridx];
utils::Assert(root_index[ridx] < (unsigned)tree.param.num_roots, "root index exceed setting");
utils::Assert(root_index[ridx] < (unsigned)tree.param.num_roots,
"root index exceed setting");
}
}
// mark delete for the deleted datas
@ -154,11 +160,12 @@ class ColMaker: public IUpdater {
}
unsigned n = static_cast<unsigned>(param.colsample_bytree * feat_index.size());
random::Shuffle(feat_index);
//utils::Check(n > 0, "colsample_bytree is too small that no feature can be included");
utils::Check(n > 0, "colsample_bytree=%g is too small that no feature can be included", param.colsample_bytree);
utils::Check(n > 0, "colsample_bytree=%g is too small that no feature can be included",
param.colsample_bytree);
feat_index.resize(n);
}
{// setup temp space for each thread
{
// setup temp space for each thread
#pragma omp parallel
{
this->nthread = omp_get_num_threads();
@ -171,20 +178,25 @@ class ColMaker: public IUpdater {
}
snode.reserve(256);
}
{// expand query
{
// expand query
qexpand_.reserve(256); qexpand_.clear();
for (int i = 0; i < tree.param.num_roots; ++i) {
qexpand_.push_back(i);
}
}
}
/*! \brief initialize the base_weight, root_gain, and NodeEntry for all the new nodes in qexpand */
/*!
* \brief initialize the base_weight, root_gain,
* and NodeEntry for all the new nodes in qexpand
*/
inline void InitNewNode(const std::vector<int> &qexpand,
const std::vector<bst_gpair> &gpair,
const IFMatrix &fmat,
const BoosterInfo &info,
const RegTree &tree) {
{// setup statistics space for each tree node
{
// setup statistics space for each tree node
for (size_t i = 0; i < stemp.size(); ++i) {
stemp[i].resize(tree.param.num_nodes, ThreadEntry(param));
}
@ -226,7 +238,7 @@ class ColMaker: public IUpdater {
}
// use new nodes for qexpand
qexpand = newnodes;
}
}
// parallel find the best split of current fid
// this function does not support nested functions
inline void ParallelFindSplit(const ColBatch::Inst &col,
@ -280,26 +292,30 @@ class ColMaker: public IUpdater {
ThreadEntry &e = stemp[tid][nid];
float fsplit;
if (tid != 0) {
if(std::abs(stemp[tid - 1][nid].last_fvalue - e.first_fvalue) > rt_2eps) {
if (std::abs(stemp[tid - 1][nid].last_fvalue - e.first_fvalue) > rt_2eps) {
fsplit = (stemp[tid - 1][nid].last_fvalue - e.first_fvalue) * 0.5f;
} else {
continue;
}
} else {
fsplit = e.first_fvalue - rt_eps;
}
}
if (need_forward && tid != 0) {
c.SetSubstract(snode[nid].stats, e.stats);
if (c.sum_hess >= param.min_child_weight && e.stats.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) + c.CalcGain(param) - snode[nid].root_gain);
if (c.sum_hess >= param.min_child_weight &&
e.stats.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) +
c.CalcGain(param) - snode[nid].root_gain);
e.best.Update(loss_chg, fid, fsplit, false);
}
}
if (need_backward) {
tmp.SetSubstract(sum, e.stats);
c.SetSubstract(snode[nid].stats, tmp);
if (c.sum_hess >= param.min_child_weight && tmp.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(tmp.CalcGain(param) + c.CalcGain(param) - snode[nid].root_gain);
if (c.sum_hess >= param.min_child_weight &&
tmp.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(tmp.CalcGain(param) +
c.CalcGain(param) - snode[nid].root_gain);
e.best.Update(loss_chg, fid, fsplit, true);
}
}
@ -308,8 +324,10 @@ class ColMaker: public IUpdater {
tmp = sum;
ThreadEntry &e = stemp[nthread-1][nid];
c.SetSubstract(snode[nid].stats, tmp);
if (c.sum_hess >= param.min_child_weight && tmp.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(tmp.CalcGain(param) + c.CalcGain(param) - snode[nid].root_gain);
if (c.sum_hess >= param.min_child_weight &&
tmp.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(tmp.CalcGain(param) +
c.CalcGain(param) - snode[nid].root_gain);
e.best.Update(loss_chg, fid, e.last_fvalue + rt_eps, true);
}
}
@ -335,25 +353,31 @@ class ColMaker: public IUpdater {
e.first_fvalue = fvalue;
} else {
// forward default right
if (std::abs(fvalue - e.first_fvalue) > rt_2eps){
if (need_forward) {
if (std::abs(fvalue - e.first_fvalue) > rt_2eps) {
if (need_forward) {
c.SetSubstract(snode[nid].stats, e.stats);
if (c.sum_hess >= param.min_child_weight && e.stats.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) + c.CalcGain(param) - snode[nid].root_gain);
if (c.sum_hess >= param.min_child_weight &&
e.stats.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) +
c.CalcGain(param) -
snode[nid].root_gain);
e.best.Update(loss_chg, fid, (fvalue + e.first_fvalue) * 0.5f, false);
}
}
if (need_backward) {
cright.SetSubstract(e.stats_extra, e.stats);
c.SetSubstract(snode[nid].stats, cright);
if (c.sum_hess >= param.min_child_weight && cright.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(cright.CalcGain(param) + c.CalcGain(param) - snode[nid].root_gain);
if (c.sum_hess >= param.min_child_weight &&
cright.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(cright.CalcGain(param) +
c.CalcGain(param) -
snode[nid].root_gain);
e.best.Update(loss_chg, fid, (fvalue + e.first_fvalue) * 0.5f, true);
}
}
}
}
e.stats.Add(gpair, info, ridx);
e.first_fvalue = fvalue;
e.first_fvalue = fvalue;
}
}
}
@ -361,7 +385,7 @@ class ColMaker: public IUpdater {
// update enumeration solution
inline void UpdateEnumeration(int nid, bst_gpair gstats,
float fvalue, int d_step, bst_uint fid,
TStats &c, std::vector<ThreadEntry> &temp) {
TStats &c, std::vector<ThreadEntry> &temp) { // NOLINT(*)
// get the statistics of nid
ThreadEntry &e = temp[nid];
// test if first hit, this is fine, because we set 0 during init
@ -370,10 +394,12 @@ class ColMaker: public IUpdater {
e.last_fvalue = fvalue;
} else {
// try to find a split
if (std::abs(fvalue - e.last_fvalue) > rt_2eps && e.stats.sum_hess >= param.min_child_weight) {
if (std::abs(fvalue - e.last_fvalue) > rt_2eps &&
e.stats.sum_hess >= param.min_child_weight) {
c.SetSubstract(snode[nid].stats, e.stats);
if (c.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) + c.CalcGain(param) - snode[nid].root_gain);
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) +
c.CalcGain(param) - snode[nid].root_gain);
e.best.Update(loss_chg, fid, (fvalue + e.last_fvalue) * 0.5f, d_step == -1);
}
}
@ -388,7 +414,7 @@ class ColMaker: public IUpdater {
int d_step,
bst_uint fid,
const std::vector<bst_gpair> &gpair,
std::vector<ThreadEntry> &temp) {
std::vector<ThreadEntry> &temp) { // NOLINT(*)
const std::vector<int> &qexpand = qexpand_;
// clear all the temp statistics
for (size_t j = 0; j < qexpand.size(); ++j) {
@ -423,7 +449,7 @@ class ColMaker: public IUpdater {
this->UpdateEnumeration(nid, buf_gpair[i],
p->fvalue, d_step,
fid, c, temp);
}
}
}
// finish up the ending piece
for (it = align_end, i = 0; it != end; ++i, it += d_step) {
@ -436,14 +462,15 @@ class ColMaker: public IUpdater {
this->UpdateEnumeration(nid, buf_gpair[i],
it->fvalue, d_step,
fid, c, temp);
}
}
// finish updating all statistics, check if it is possible to include all sum statistics
for (size_t i = 0; i < qexpand.size(); ++i) {
const int nid = qexpand[i];
ThreadEntry &e = temp[nid];
c.SetSubstract(snode[nid].stats, e.stats);
if (e.stats.sum_hess >= param.min_child_weight && c.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) + c.CalcGain(param) - snode[nid].root_gain);
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) +
c.CalcGain(param) - snode[nid].root_gain);
const float gap = std::abs(e.last_fvalue) + rt_eps;
const float delta = d_step == +1 ? gap: -gap;
e.best.Update(loss_chg, fid, e.last_fvalue + delta, d_step == -1);
@ -458,7 +485,7 @@ class ColMaker: public IUpdater {
bst_uint fid,
const std::vector<bst_gpair> &gpair,
const BoosterInfo &info,
std::vector<ThreadEntry> &temp) {
std::vector<ThreadEntry> &temp) { // NOLINT(*)
// use cacheline aware optimization
if (TStats::kSimpleStats != 0 && param.cache_opt != 0) {
EnumerateSplitCacheOpt(begin, end, d_step, fid, gpair, temp);
@ -471,7 +498,7 @@ class ColMaker: public IUpdater {
}
// left statistics
TStats c(param);
for(const ColBatch::Entry *it = begin; it != end; it += d_step) {
for (const ColBatch::Entry *it = begin; it != end; it += d_step) {
const bst_uint ridx = it->index;
const int nid = position[ridx];
if (nid < 0) continue;
@ -485,10 +512,12 @@ class ColMaker: public IUpdater {
e.last_fvalue = fvalue;
} else {
// try to find a split
if (std::abs(fvalue - e.last_fvalue) > rt_2eps && e.stats.sum_hess >= param.min_child_weight) {
if (std::abs(fvalue - e.last_fvalue) > rt_2eps &&
e.stats.sum_hess >= param.min_child_weight) {
c.SetSubstract(snode[nid].stats, e.stats);
if (c.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) + c.CalcGain(param) - snode[nid].root_gain);
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) +
c.CalcGain(param) - snode[nid].root_gain);
e.best.Update(loss_chg, fid, (fvalue + e.last_fvalue) * 0.5f, d_step == -1);
}
}
@ -503,7 +532,8 @@ class ColMaker: public IUpdater {
ThreadEntry &e = temp[nid];
c.SetSubstract(snode[nid].stats, e.stats);
if (e.stats.sum_hess >= param.min_child_weight && c.sum_hess >= param.min_child_weight) {
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) + c.CalcGain(param) - snode[nid].root_gain);
bst_float loss_chg = static_cast<bst_float>(e.stats.CalcGain(param) +
c.CalcGain(param) - snode[nid].root_gain);
const float gap = std::abs(e.last_fvalue) + rt_eps;
const float delta = d_step == +1 ? gap: -gap;
e.best.Update(loss_chg, fid, e.last_fvalue + delta, d_step == -1);
@ -511,14 +541,14 @@ class ColMaker: public IUpdater {
}
}
// update the solution candidate
// update the solution candidate
virtual void UpdateSolution(const ColBatch &batch,
const std::vector<bst_gpair> &gpair,
const IFMatrix &fmat,
const BoosterInfo &info) {
// start enumeration
const bst_omp_uint nsize = static_cast<bst_omp_uint>(batch.size);
#if defined(_OPENMP)
#if defined(_OPENMP)
const int batch_size = std::max(static_cast<int>(nsize / this->nthread / 32), 1);
#endif
int poption = param.parallel_option;
@ -533,11 +563,11 @@ class ColMaker: public IUpdater {
const ColBatch::Inst c = batch[i];
const bool ind = c.length != 0 && c.data[0].fvalue == c.data[c.length - 1].fvalue;
if (param.need_forward_search(fmat.GetColDensity(fid), ind)) {
this->EnumerateSplit(c.data, c.data + c.length, +1,
this->EnumerateSplit(c.data, c.data + c.length, +1,
fid, gpair, info, stemp[tid]);
}
if (param.need_backward_search(fmat.GetColDensity(fid), ind)) {
this->EnumerateSplit(c.data + c.length - 1, c.data - 1, -1,
this->EnumerateSplit(c.data + c.length - 1, c.data - 1, -1,
fid, gpair, info, stemp[tid]);
}
}
@ -546,7 +576,7 @@ class ColMaker: public IUpdater {
this->ParallelFindSplit(batch[i], batch.col_index[i],
fmat, gpair, info);
}
}
}
}
// find splits at current level, do split per level
inline void FindSplit(int depth,
@ -571,7 +601,7 @@ class ColMaker: public IUpdater {
// get the best result, we can synchronize the solution
for (size_t i = 0; i < qexpand.size(); ++i) {
const int nid = qexpand[i];
NodeEntry &e = snode[nid];
NodeEntry &e = snode[nid];
// now we know the solution in snode[nid], set split
if (e.best.loss_chg > rt_eps) {
p_tree->AddChilds(nid);
@ -582,19 +612,20 @@ class ColMaker: public IUpdater {
} else {
(*p_tree)[nid].set_leaf(e.weight * param.learning_rate);
}
}
}
}
// reset position of each data points after split is created in the tree
inline void ResetPosition(const std::vector<int> &qexpand, IFMatrix *p_fmat, const RegTree &tree) {
inline void ResetPosition(const std::vector<int> &qexpand,
IFMatrix *p_fmat, const RegTree &tree) {
// set the positions in the nondefault
this->SetNonDefaultPosition(qexpand, p_fmat, tree);
this->SetNonDefaultPosition(qexpand, p_fmat, tree);
// set rest of instances to default position
const std::vector<bst_uint> &rowset = p_fmat->buffered_rowset();
// set default direct nodes to default
// for leaf nodes that are not fresh, mark then to ~nid,
// for leaf nodes that are not fresh, mark then to ~nid,
// so that they are ignored in future statistics collection
const bst_omp_uint ndata = static_cast<bst_omp_uint>(rowset.size());
#pragma omp parallel for schedule(static)
for (bst_omp_uint i = 0; i < ndata; ++i) {
const bst_uint ridx = rowset[i];
@ -655,7 +686,7 @@ class ColMaker: public IUpdater {
const float fvalue = col[j].fvalue;
// go back to parent, correct those who are not default
if (!tree[nid].is_leaf() && tree[nid].split_index() == fid) {
if(fvalue < tree[nid].split_cond()) {
if (fvalue < tree[nid].split_cond()) {
this->SetEncodePosition(ridx, tree[nid].cleft());
} else {
this->SetEncodePosition(ridx, tree[nid].cright());
@ -667,7 +698,7 @@ class ColMaker: public IUpdater {
}
// utils to get/set position, with encoded format
// return decoded position
inline int DecodePosition(bst_uint ridx) const{
inline int DecodePosition(bst_uint ridx) const {
const int pid = position[ridx];
return pid < 0 ? ~pid : pid;
}
@ -679,7 +710,7 @@ class ColMaker: public IUpdater {
position[ridx] = nid;
}
}
//--data fields--
// --data fields--
const TrainParam &param;
// number of omp thread used during training
int nthread;

32
src/tree/updater_distcol-inl.hpp
View File

@ -1,11 +1,15 @@
#ifndef XGBOOST_TREE_UPDATER_DISTCOL_INL_HPP_
#define XGBOOST_TREE_UPDATER_DISTCOL_INL_HPP_
/*!
* Copyright 2014 by Contributors
* \file updater_distcol-inl.hpp
* \brief beta distributed version that takes a sub-column
* \brief beta distributed version that takes a sub-column
* and construct a tree
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_UPDATER_DISTCOL_INL_HPP_
#define XGBOOST_TREE_UPDATER_DISTCOL_INL_HPP_
#include <vector>
#include <algorithm>
#include "../sync/sync.h"
#include "../utils/bitmap.h"
#include "../utils/io.h"
@ -27,7 +31,7 @@ class DistColMaker : public ColMaker<TStats> {
virtual void Update(const std::vector<bst_gpair> &gpair,
IFMatrix *p_fmat,
const BoosterInfo &info,
const std::vector<RegTree*> &trees) {
const std::vector<RegTree*> &trees) {
TStats::CheckInfo(info);
utils::Check(trees.size() == 1, "DistColMaker: only support one tree at a time");
// build the tree
@ -39,11 +43,12 @@ class DistColMaker : public ColMaker<TStats> {
}
virtual const int* GetLeafPosition(void) const {
return builder.GetLeafPosition();
}
}
private:
struct Builder : public ColMaker<TStats>::Builder {
public:
Builder(const TrainParam &param)
explicit Builder(const TrainParam &param)
: ColMaker<TStats>::Builder(param) {
}
inline void UpdatePosition(IFMatrix *p_fmat, const RegTree &tree) {
@ -63,7 +68,8 @@ class DistColMaker : public ColMaker<TStats> {
virtual const int* GetLeafPosition(void) const {
return BeginPtr(this->position);
}
protected:
protected:
virtual void SetNonDefaultPosition(const std::vector<int> &qexpand,
IFMatrix *p_fmat, const RegTree &tree) {
// step 2, classify the non-default data into right places
@ -87,7 +93,7 @@ class DistColMaker : public ColMaker<TStats> {
#pragma omp parallel for schedule(static)
for (bst_omp_uint j = 0; j < ndata; ++j) {
boolmap[j] = 0;
}
}
}
utils::IIterator<ColBatch> *iter = p_fmat->ColIterator(fsplits);
while (iter->Next()) {
@ -111,7 +117,7 @@ class DistColMaker : public ColMaker<TStats> {
}
}
}
bitmap.InitFromBool(boolmap);
// communicate bitmap
rabit::Allreduce<rabit::op::BitOR>(BeginPtr(bitmap.data), bitmap.data.size());
@ -142,7 +148,7 @@ class DistColMaker : public ColMaker<TStats> {
}
vec.push_back(this->snode[nid].best);
}
// TODO, lazy version
// TODO(tqchen) lazy version
// communicate best solution
reducer.Allreduce(BeginPtr(vec), vec.size());
// assign solution back
@ -151,7 +157,7 @@ class DistColMaker : public ColMaker<TStats> {
this->snode[nid].best = vec[i];
}
}
private:
utils::BitMap bitmap;
std::vector<int> boolmap;
@ -162,8 +168,8 @@ class DistColMaker : public ColMaker<TStats> {
// training parameter
TrainParam param;
// pointer to the builder
Builder builder;
Builder builder;
};
} // namespace tree
} // namespace xgboost
#endif
#endif // XGBOOST_TREE_UPDATER_DISTCOL_INL_HPP_

92
src/tree/updater_histmaker-inl.hpp
View File

@ -1,10 +1,12 @@
#ifndef XGBOOST_TREE_UPDATER_HISTMAKER_INL_HPP_
#define XGBOOST_TREE_UPDATER_HISTMAKER_INL_HPP_
/*!
* Copyright 2014 by Contributors
* \file updater_histmaker-inl.hpp
* \brief use histogram counting to construct a tree
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_UPDATER_HISTMAKER_INL_HPP_
#define XGBOOST_TREE_UPDATER_HISTMAKER_INL_HPP_
#include <vector>
#include <algorithm>
#include "../sync/sync.h"
@ -38,7 +40,7 @@ class HistMaker: public BaseMaker {
struct HistUnit {
/*! \brief cutting point of histogram, contains maximum point */
const bst_float *cut;
/*! \brief content of statistics data */
/*! \brief content of statistics data */
TStats *data;
/*! \brief size of histogram */
unsigned size;
@ -48,13 +50,13 @@ class HistMaker: public BaseMaker {
HistUnit(const bst_float *cut, TStats *data, unsigned size)
: cut(cut), data(data), size(size) {}
/*! \brief add a histogram to data */
inline void Add(bst_float fv,
inline void Add(bst_float fv,
const std::vector<bst_gpair> &gpair,
const BoosterInfo &info,
const bst_uint ridx) {
unsigned i = std::upper_bound(cut, cut + size, fv) - cut;
utils::Assert(size != 0, "try insert into size=0");
utils::Assert(i < size,
utils::Assert(i < size,
"maximum value must be in cut, fv = %g, cutmax=%g", fv, cut[size-1]);
data[i].Add(gpair, info, ridx);
}
@ -74,7 +76,7 @@ class HistMaker: public BaseMaker {
rptr[fid+1] - rptr[fid]);
}
};
// thread workspace
// thread workspace
struct ThreadWSpace {
/*! \brief actual unit pointer */
std::vector<unsigned> rptr;
@ -92,7 +94,7 @@ class HistMaker: public BaseMaker {
}
hset[tid].rptr = BeginPtr(rptr);
hset[tid].cut = BeginPtr(cut);
hset[tid].data.resize(cut.size(), TStats(param));
hset[tid].data.resize(cut.size(), TStats(param));
}
}
// aggregate all statistics to hset[0]
@ -147,7 +149,7 @@ class HistMaker: public BaseMaker {
}
// this function does two jobs
// (1) reset the position in array position, to be the latest leaf id
// (2) propose a set of candidate cuts and set wspace.rptr wspace.cut correctly
// (2) propose a set of candidate cuts and set wspace.rptr wspace.cut correctly
virtual void ResetPosAndPropose(const std::vector<bst_gpair> &gpair,
IFMatrix *p_fmat,
const BoosterInfo &info,
@ -171,8 +173,9 @@ class HistMaker: public BaseMaker {
const BoosterInfo &info,
const std::vector <bst_uint> &fset,
const RegTree &tree) = 0;
private:
inline void EnumerateSplit(const HistUnit &hist,
inline void EnumerateSplit(const HistUnit &hist,
const TStats &node_sum,
bst_uint fid,
SplitEntry *best,
@ -187,7 +190,7 @@ class HistMaker: public BaseMaker {
c.SetSubstract(node_sum, s);
if (c.sum_hess >= param.min_child_weight) {
double loss_chg = s.CalcGain(param) + c.CalcGain(param) - root_gain;
if (best->Update((float)loss_chg, fid, hist.cut[i], false)) {
if (best->Update(static_cast<float>(loss_chg), fid, hist.cut[i], false)) {
*left_sum = s;
}
}
@ -200,7 +203,7 @@ class HistMaker: public BaseMaker {
c.SetSubstract(node_sum, s);
if (c.sum_hess >= param.min_child_weight) {
double loss_chg = s.CalcGain(param) + c.CalcGain(param) - root_gain;
if (best->Update((float)loss_chg, fid, hist.cut[i-1], true)) {
if (best->Update(static_cast<float>(loss_chg), fid, hist.cut[i-1], true)) {
*left_sum = c;
}
}
@ -216,22 +219,22 @@ class HistMaker: public BaseMaker {
const size_t num_feature = fset.size();
// get the best split condition for each node
std::vector<SplitEntry> sol(qexpand.size());
std::vector<TStats> left_sum(qexpand.size());
std::vector<TStats> left_sum(qexpand.size());
bst_omp_uint nexpand = static_cast<bst_omp_uint>(qexpand.size());
#pragma omp parallel for schedule(dynamic, 1)
for (bst_omp_uint wid = 0; wid < nexpand; ++ wid) {
for (bst_omp_uint wid = 0; wid < nexpand; ++wid) {
const int nid = qexpand[wid];
utils::Assert(node2workindex[nid] == static_cast<int>(wid),
"node2workindex inconsistent");
SplitEntry &best = sol[wid];
TStats &node_sum = wspace.hset[0][num_feature + wid * (num_feature + 1)].data[0];
for (size_t i = 0; i < fset.size(); ++ i) {
for (size_t i = 0; i < fset.size(); ++i) {
EnumerateSplit(this->wspace.hset[0][i + wid * (num_feature+1)],
node_sum, fset[i], &best, &left_sum[wid]);
}
}
// get the best result, we can synchronize the solution
for (bst_omp_uint wid = 0; wid < nexpand; ++ wid) {
for (bst_omp_uint wid = 0; wid < nexpand; ++wid) {
const int nid = qexpand[wid];
const SplitEntry &best = sol[wid];
const TStats &node_sum = wspace.hset[0][num_feature + wid * (num_feature + 1)].data[0];
@ -244,7 +247,7 @@ class HistMaker: public BaseMaker {
(*p_tree)[nid].set_split(best.split_index(),
best.split_value, best.default_left());
// mark right child as 0, to indicate fresh leaf
(*p_tree)[(*p_tree)[nid].cleft()].set_leaf(0.0f, 0);
(*p_tree)[(*p_tree)[nid].cleft()].set_leaf(0.0f, 0);
(*p_tree)[(*p_tree)[nid].cright()].set_leaf(0.0f, 0);
// right side sum
TStats right_sum;
@ -256,11 +259,11 @@ class HistMaker: public BaseMaker {
}
}
}
inline void SetStats(RegTree *p_tree, int nid, const TStats &node_sum) {
p_tree->stat(nid).base_weight = static_cast<float>(node_sum.CalcWeight(param));
p_tree->stat(nid).sum_hess = static_cast<float>(node_sum.sum_hess);
node_sum.SetLeafVec(param, p_tree->leafvec(nid));
node_sum.SetLeafVec(param, p_tree->leafvec(nid));
}
};
@ -270,7 +273,7 @@ class CQHistMaker: public HistMaker<TStats> {
struct HistEntry {
typename HistMaker<TStats>::HistUnit hist;
unsigned istart;
/*!
/*!
* \brief add a histogram to data,
* do linear scan, start from istart
*/
@ -282,7 +285,7 @@ class CQHistMaker: public HistMaker<TStats> {
utils::Assert(istart != hist.size, "the bound variable must be max");
hist.data[istart].Add(gpair, info, ridx);
}
/*!
/*!
* \brief add a histogram to data,
* do linear scan, start from istart
*/
@ -302,7 +305,7 @@ class CQHistMaker: public HistMaker<TStats> {
feat_helper.InitByCol(p_fmat, tree);
feat_helper.SampleCol(this->param.colsample_bytree, p_fset);
}
// code to create histogram
// code to create histogram
virtual void CreateHist(const std::vector<bst_gpair> &gpair,
IFMatrix *p_fmat,
const BoosterInfo &info,
@ -313,7 +316,7 @@ class CQHistMaker: public HistMaker<TStats> {
std::fill(feat2workindex.begin(), feat2workindex.end(), -1);
for (size_t i = 0; i < fset.size(); ++i) {
feat2workindex[fset[i]] = static_cast<int>(i);
}
}
// start to work
this->wspace.Init(this->param, 1);
// if it is C++11, use lazy evaluation for Allreduce,
@ -350,11 +353,11 @@ class CQHistMaker: public HistMaker<TStats> {
// sync the histogram
// if it is C++11, use lazy evaluation for Allreduce
#if __cplusplus >= 201103L
this->histred.Allreduce(BeginPtr(this->wspace.hset[0].data),
this->histred.Allreduce(BeginPtr(this->wspace.hset[0].data),
this->wspace.hset[0].data.size(), lazy_get_hist);
#else
this->histred.Allreduce(BeginPtr(this->wspace.hset[0].data), this->wspace.hset[0].data.size());
#endif
this->histred.Allreduce(BeginPtr(this->wspace.hset[0].data), this->wspace.hset[0].data.size());
#endif
}
virtual void ResetPositionAfterSplit(IFMatrix *p_fmat,
const RegTree &tree) {
@ -374,11 +377,11 @@ class CQHistMaker: public HistMaker<TStats> {
feat2workindex[fset[i]] = static_cast<int>(freal_set.size());
freal_set.push_back(fset[i]);
} else {
feat2workindex[fset[i]] = -2;
feat2workindex[fset[i]] = -2;
}
}
this->GetNodeStats(gpair, *p_fmat, tree, info,
&thread_stats, &node_stats);
&thread_stats, &node_stats);
sketchs.resize(this->qexpand.size() * freal_set.size());
for (size_t i = 0; i < sketchs.size(); ++i) {
sketchs[i].Init(info.num_row, this->param.sketch_eps);
@ -394,7 +397,8 @@ class CQHistMaker: public HistMaker<TStats> {
#if __cplusplus >= 201103L
auto lazy_get_summary = [&]()
#endif
{// get smmary
{
// get smmary
thread_sketch.resize(this->get_nthread());
// number of rows in
const size_t nrows = p_fmat->buffered_rowset().size();
@ -457,9 +461,9 @@ class CQHistMaker: public HistMaker<TStats> {
this->wspace.rptr.push_back(static_cast<unsigned>(this->wspace.cut.size()));
} else {
utils::Assert(offset == -2, "BUG in mark");
bst_float cpt = feat_helper.MaxValue(fset[i]);
bst_float cpt = feat_helper.MaxValue(fset[i]);
this->wspace.cut.push_back(cpt + fabs(cpt) + rt_eps);
this->wspace.rptr.push_back(static_cast<unsigned>(this->wspace.cut.size()));
this->wspace.rptr.push_back(static_cast<unsigned>(this->wspace.cut.size()));
}
}
// reserve last value for global statistics
@ -470,7 +474,7 @@ class CQHistMaker: public HistMaker<TStats> {
(fset.size() + 1) * this->qexpand.size() + 1,
"cut space inconsistent");
}
private:
inline void UpdateHistCol(const std::vector<bst_gpair> &gpair,
const ColBatch::Inst &c,
@ -554,9 +558,9 @@ class CQHistMaker: public HistMaker<TStats> {
}
} else {
for (size_t i = 0; i < this->qexpand.size(); ++i) {
const unsigned nid = this->qexpand[i];
const unsigned nid = this->qexpand[i];
sbuilder[nid].sum_total = static_cast<bst_float>(nstats[nid].sum_hess);
}
}
}
// if only one value, no need to do second pass
if (c[0].fvalue == c[c.length-1].fvalue) {
@ -589,7 +593,7 @@ class CQHistMaker: public HistMaker<TStats> {
if (nid >= 0) {
sbuilder[nid].Push(c[j + i].fvalue, buf_hess[i], max_size);
}
}
}
}
for (bst_uint j = align_length; j < c.length; ++j) {
const bst_uint ridx = c[j].index;
@ -617,7 +621,7 @@ class CQHistMaker: public HistMaker<TStats> {
// temp space to map feature id to working index
std::vector<int> feat2workindex;
// set of index from fset that are real
std::vector<bst_uint> freal_set;
std::vector<bst_uint> freal_set;
// thread temp data
std::vector< std::vector<BaseMaker::SketchEntry> > thread_sketch;
// used to hold statistics
@ -631,18 +635,18 @@ class CQHistMaker: public HistMaker<TStats> {
// reducer for summary
rabit::SerializeReducer<WXQSketch::SummaryContainer> sreducer;
// per node, per feature sketch
std::vector< utils::WXQuantileSketch<bst_float, bst_float> > sketchs;
std::vector< utils::WXQuantileSketch<bst_float, bst_float> > sketchs;
};
template<typename TStats>
class QuantileHistMaker: public HistMaker<TStats> {
class QuantileHistMaker: public HistMaker<TStats> {
protected:
typedef utils::WXQuantileSketch<bst_float, bst_float> WXQSketch;
virtual void ResetPosAndPropose(const std::vector<bst_gpair> &gpair,
IFMatrix *p_fmat,
const BoosterInfo &info,
const std::vector <bst_uint> &fset,
const RegTree &tree) {
const RegTree &tree) {
// initialize the data structure
int nthread = BaseMaker::get_nthread();
sketchs.resize(this->qexpand.size() * tree.param.num_feature);
@ -658,7 +662,7 @@ class QuantileHistMaker: public HistMaker<TStats> {
utils::ParallelGroupBuilder<SparseBatch::Entry> builder(&col_ptr, &col_data, &thread_col_ptr);
builder.InitBudget(tree.param.num_feature, nthread);
const bst_omp_uint nbatch = static_cast<bst_omp_uint>(batch.size);
const bst_omp_uint nbatch = static_cast<bst_omp_uint>(batch.size);
#pragma omp parallel for schedule(static)
for (bst_omp_uint i = 0; i < nbatch; ++i) {
RowBatch::Inst inst = batch[i];
@ -667,11 +671,11 @@ class QuantileHistMaker: public HistMaker<TStats> {
if (nid >= 0) {
if (!tree[nid].is_leaf()) {
this->position[ridx] = nid = HistMaker<TStats>::NextLevel(inst, tree, nid);
}
}
if (this->node2workindex[nid] < 0) {
this->position[ridx] = ~nid;
} else{
for (bst_uint j = 0; j < inst.length; ++j) {
} else {
for (bst_uint j = 0; j < inst.length; ++j) {
builder.AddBudget(inst[j].index, omp_get_thread_num());
}
}
@ -712,8 +716,8 @@ class QuantileHistMaker: public HistMaker<TStats> {
summary_array[i].Reserve(max_size);
summary_array[i].SetPrune(out, max_size);
}
size_t nbytes = WXQSketch::SummaryContainer::CalcMemCost(max_size);
size_t nbytes = WXQSketch::SummaryContainer::CalcMemCost(max_size);
sreducer.Allreduce(BeginPtr(summary_array), nbytes, summary_array.size());
// now we get the final result of sketch, setup the cut
this->wspace.cut.clear();

15
src/tree/updater_prune-inl.hpp
View File

@ -1,10 +1,12 @@
#ifndef XGBOOST_TREE_UPDATER_PRUNE_INL_HPP_
#define XGBOOST_TREE_UPDATER_PRUNE_INL_HPP_
/*!
* Copyright 2014 by Contributors
* \file updater_prune-inl.hpp
* \brief prune a tree given the statistics
* \brief prune a tree given the statistics
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_UPDATER_PRUNE_INL_HPP_
#define XGBOOST_TREE_UPDATER_PRUNE_INL_HPP_
#include <vector>
#include "./param.h"
#include "./updater.h"
@ -37,9 +39,10 @@ class TreePruner: public IUpdater {
param.learning_rate = lr;
syncher.Update(gpair, p_fmat, info, trees);
}
private:
// try to prune off current leaf
inline int TryPruneLeaf(RegTree &tree, int nid, int depth, int npruned) {
inline int TryPruneLeaf(RegTree &tree, int nid, int depth, int npruned) { // NOLINT(*)
if (tree[nid].is_root()) return npruned;
int pid = tree[nid].parent();
RegTree::NodeStat &s = tree.stat(pid);
@ -51,10 +54,10 @@ class TreePruner: public IUpdater {
return this->TryPruneLeaf(tree, pid, depth - 1, npruned+2);
} else {
return npruned;
}
}
}
/*! \brief do prunning of a tree */
inline void DoPrune(RegTree &tree) {
inline void DoPrune(RegTree &tree) { // NOLINT(*)
int npruned = 0;
// initialize auxiliary statistics
for (int nid = 0; nid < tree.param.num_nodes; ++nid) {

12
src/tree/updater_refresh-inl.hpp
View File

@ -1,10 +1,12 @@
#ifndef XGBOOST_TREE_UPDATER_REFRESH_INL_HPP_
#define XGBOOST_TREE_UPDATER_REFRESH_INL_HPP_
/*!
* Copyright 2014 by Contributors
* \file updater_refresh-inl.hpp
* \brief refresh the statistics and leaf value on the tree on the dataset
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_UPDATER_REFRESH_INL_HPP_
#define XGBOOST_TREE_UPDATER_REFRESH_INL_HPP_
#include <vector>
#include <limits>
#include "../sync/sync.h"
@ -27,7 +29,7 @@ class TreeRefresher: public IUpdater {
virtual void Update(const std::vector<bst_gpair> &gpair,
IFMatrix *p_fmat,
const BoosterInfo &info,
const std::vector<RegTree*> &trees) {
const std::vector<RegTree*> &trees) {
if (trees.size() == 0) return;
// number of threads
// thread temporal space
@ -100,7 +102,7 @@ class TreeRefresher: public IUpdater {
float lr = param.learning_rate;
param.learning_rate = lr / trees.size();
int offset = 0;
for (size_t i = 0; i < trees.size(); ++i) {
for (size_t i = 0; i < trees.size(); ++i) {
for (int rid = 0; rid < trees[i]->param.num_roots; ++rid) {
this->Refresh(BeginPtr(stemp[0]) + offset, rid, trees[i]);
}
@ -147,7 +149,7 @@ class TreeRefresher: public IUpdater {
// training parameter
TrainParam param;
// reducer
rabit::Reducer<TStats, TStats::Reduce> reducer;
rabit::Reducer<TStats, TStats::Reduce> reducer;
};
} // namespace tree

59
src/tree/updater_skmaker-inl.hpp
View File

@ -1,11 +1,13 @@
#ifndef XGBOOST_TREE_UPDATER_SKMAKER_INL_HPP_
#define XGBOOST_TREE_UPDATER_SKMAKER_INL_HPP_
/*!
* Copyright 2014 by Contributors
* \file updater_skmaker-inl.hpp
* \brief use approximation sketch to construct a tree,
a refresh is needed to make the statistics exactly correct
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_UPDATER_SKMAKER_INL_HPP_
#define XGBOOST_TREE_UPDATER_SKMAKER_INL_HPP_
#include <vector>
#include <algorithm>
#include "../sync/sync.h"
@ -30,7 +32,7 @@ class SketchMaker: public BaseMaker {
}
param.learning_rate = lr;
}
protected:
inline void Update(const std::vector<bst_gpair> &gpair,
IFMatrix *p_fmat,
@ -79,9 +81,9 @@ class SketchMaker: public BaseMaker {
double pos_grad;
/*! \brief sum of all negative gradient */
double neg_grad;
/*! \brief sum of hessian statistics */
/*! \brief sum of hessian statistics */
double sum_hess;
explicit SKStats(void) {}
SKStats(void) {}
// constructor
explicit SKStats(const TrainParam &param) {
this->Clear();
@ -123,7 +125,7 @@ class SketchMaker: public BaseMaker {
sum_hess += b.sum_hess;
}
/*! \brief same as add, reduce is used in All Reduce */
inline static void Reduce(SKStats &a, const SKStats &b) {
inline static void Reduce(SKStats &a, const SKStats &b) { // NOLINT(*)
a.Add(b);
}
/*! \brief set leaf vector value based on statistics */
@ -139,7 +141,7 @@ class SketchMaker: public BaseMaker {
sketchs[i].Init(info.num_row, this->param.sketch_eps);
}
thread_sketch.resize(this->get_nthread());
// number of rows in
// number of rows in
const size_t nrows = p_fmat->buffered_rowset().size();
// start accumulating statistics
utils::IIterator<ColBatch> *iter = p_fmat->ColIterator();
@ -156,7 +158,7 @@ class SketchMaker: public BaseMaker {
batch[i].length == nrows,
&thread_sketch[omp_get_thread_num()]);
}
}
}
// setup maximum size
unsigned max_size = param.max_sketch_size();
// synchronize sketch
@ -167,8 +169,8 @@ class SketchMaker: public BaseMaker {
summary_array[i].Reserve(max_size);
summary_array[i].SetPrune(out, max_size);
}
size_t nbytes = WXQSketch::SummaryContainer::CalcMemCost(max_size);
sketch_reducer.Allreduce(BeginPtr(summary_array), nbytes, summary_array.size());
size_t nbytes = WXQSketch::SummaryContainer::CalcMemCost(max_size);
sketch_reducer.Allreduce(BeginPtr(summary_array), nbytes, summary_array.size());
}
// update sketch information in column fid
inline void UpdateSketchCol(const std::vector<bst_gpair> &gpair,
@ -209,7 +211,7 @@ class SketchMaker: public BaseMaker {
const unsigned nid = this->qexpand[i];
sbuilder[3 * nid + 0].sum_total = static_cast<bst_float>(nstats[nid].pos_grad);
sbuilder[3 * nid + 1].sum_total = static_cast<bst_float>(nstats[nid].neg_grad);
sbuilder[3 * nid + 2].sum_total = static_cast<bst_float>(nstats[nid].sum_hess);
sbuilder[3 * nid + 2].sum_total = static_cast<bst_float>(nstats[nid].sum_hess);
}
}
// if only one value, no need to do second pass
@ -217,7 +219,9 @@ class SketchMaker: public BaseMaker {
for (size_t i = 0; i < this->qexpand.size(); ++i) {
const int nid = this->qexpand[i];
for (int k = 0; k < 3; ++k) {
sbuilder[3 * nid + k].sketch->Push(c[0].fvalue, static_cast<bst_float>(sbuilder[3 * nid + k].sum_total));
sbuilder[3 * nid + k].sketch->Push(c[0].fvalue,
static_cast<bst_float>(
sbuilder[3 * nid + k].sum_total));
}
}
return;
@ -250,7 +254,7 @@ class SketchMaker: public BaseMaker {
sbuilder[3 * nid + k].Finalize(max_size);
}
}
}
}
inline void SyncNodeStats(void) {
utils::Assert(qexpand.size() != 0, "qexpand must not be empty");
std::vector<SKStats> tmp(qexpand.size());
@ -272,12 +276,12 @@ class SketchMaker: public BaseMaker {
std::vector<SplitEntry> sol(qexpand.size());
bst_omp_uint nexpand = static_cast<bst_omp_uint>(qexpand.size());
#pragma omp parallel for schedule(dynamic, 1)
for (bst_omp_uint wid = 0; wid < nexpand; ++ wid) {
for (bst_omp_uint wid = 0; wid < nexpand; ++wid) {
const int nid = qexpand[wid];
utils::Assert(node2workindex[nid] == static_cast<int>(wid),
"node2workindex inconsistent");
SplitEntry &best = sol[wid];
for (bst_uint fid = 0; fid < num_feature; ++ fid) {
for (bst_uint fid = 0; fid < num_feature; ++fid) {
unsigned base = (wid * p_tree->param.num_feature + fid) * 3;
EnumerateSplit(summary_array[base + 0],
summary_array[base + 1],
@ -286,7 +290,7 @@ class SketchMaker: public BaseMaker {
}
}
// get the best result, we can synchronize the solution
for (bst_omp_uint wid = 0; wid < nexpand; ++ wid) {
for (bst_omp_uint wid = 0; wid < nexpand; ++wid) {
const int nid = qexpand[wid];
const SplitEntry &best = sol[wid];
// set up the values
@ -337,7 +341,7 @@ class SketchMaker: public BaseMaker {
feat_sum.neg_grad = neg_grad.data[neg_grad.size - 1].rmax;
feat_sum.sum_hess = sum_hess.data[sum_hess.size - 1].rmax;
size_t ipos = 0, ineg = 0, ihess = 0;
for (size_t i = 1; i < fsplits.size(); ++i) {
for (size_t i = 1; i < fsplits.size(); ++i) {
WXQSketch::Entry pos = pos_grad.Query(fsplits[i], ipos);
WXQSketch::Entry neg = neg_grad.Query(fsplits[i], ineg);
WXQSketch::Entry hess = sum_hess.Query(fsplits[i], ihess);
@ -345,11 +349,11 @@ class SketchMaker: public BaseMaker {
s.pos_grad = 0.5f * (pos.rmin + pos.rmax - pos.wmin);
s.neg_grad = 0.5f * (neg.rmin + neg.rmax - neg.wmin);
s.sum_hess = 0.5f * (hess.rmin + hess.rmax - hess.wmin);
c.SetSubstract(node_sum, s);
c.SetSubstract(node_sum, s);
// forward
if (s.sum_hess >= param.min_child_weight &&
c.sum_hess >= param.min_child_weight) {
double loss_chg = s.CalcGain(param) + c.CalcGain(param) - root_gain;
double loss_chg = s.CalcGain(param) + c.CalcGain(param) - root_gain;
best->Update(static_cast<bst_float>(loss_chg), fid, fsplits[i], false);
}
// backward
@ -357,22 +361,23 @@ class SketchMaker: public BaseMaker {
s.SetSubstract(node_sum, c);
if (s.sum_hess >= param.min_child_weight &&
c.sum_hess >= param.min_child_weight) {
double loss_chg = s.CalcGain(param) + c.CalcGain(param) - root_gain;
double loss_chg = s.CalcGain(param) + c.CalcGain(param) - root_gain;
best->Update(static_cast<bst_float>(loss_chg), fid, fsplits[i], true);
}
}
}
{// all including
{
// all including
SKStats s = feat_sum, c;
c.SetSubstract(node_sum, s);
if (s.sum_hess >= param.min_child_weight &&
c.sum_hess >= param.min_child_weight) {
bst_float cpt = fsplits.back();
double loss_chg = s.CalcGain(param) + c.CalcGain(param) - root_gain;
double loss_chg = s.CalcGain(param) + c.CalcGain(param) - root_gain;
best->Update(static_cast<bst_float>(loss_chg), fid, cpt + fabsf(cpt) + 1.0f, false);
}
}
}
// thread temp data
// used to hold temporal sketch
std::vector< std::vector<SketchEntry> > thread_sketch;
@ -389,6 +394,6 @@ class SketchMaker: public BaseMaker {
// per node, per feature sketch
std::vector< utils::WXQuantileSketch<bst_float, bst_float> > sketchs;
};
} // tree
} // xgboost
#endif
} // namespace tree
} // namespace xgboost
#endif // XGBOOST_TREE_UPDATER_SKMAKER_INL_HPP_

13
src/tree/updater_sync-inl.hpp
View File

@ -1,18 +1,21 @@
#ifndef XGBOOST_TREE_UPDATER_SYNC_INL_HPP_
#define XGBOOST_TREE_UPDATER_SYNC_INL_HPP_
/*!
* Copyright 2014 by Contributors
* \file updater_sync-inl.hpp
* \brief synchronize the tree in all distributed nodes
* \author Tianqi Chen
*/
#ifndef XGBOOST_TREE_UPDATER_SYNC_INL_HPP_
#define XGBOOST_TREE_UPDATER_SYNC_INL_HPP_
#include <vector>
#include <string>
#include <limits>
#include "../sync/sync.h"
#include "./updater.h"
namespace xgboost {
namespace tree {
/*!
/*!
* \brief syncher that synchronize the tree in all distributed nodes
* can implement various strategies, so far it is always set to node 0's tree
*/
@ -28,7 +31,7 @@ class TreeSyncher: public IUpdater {
const std::vector<RegTree*> &trees) {
this->SyncTrees(trees);
}
private:
// synchronize the trees in different nodes, take tree from rank 0
inline void SyncTrees(const std::vector<RegTree *> &trees) {
@ -43,7 +46,7 @@ class TreeSyncher: public IUpdater {
}
fs.Seek(0);
rabit::Broadcast(&s_model, 0);
for (size_t i = 0; i < trees.size(); ++i) {
for (size_t i = 0; i < trees.size(); ++i) {
trees[i]->LoadModel(fs);
}
}

32
src/utils/config.h
View File

@ -1,10 +1,12 @@
#ifndef XGBOOST_UTILS_CONFIG_H_
#define XGBOOST_UTILS_CONFIG_H_
/*!
* Copyright 2014 by Contributors
* \file config.h
* \brief helper class to load in configures from file
* \author Tianqi Chen
*/
#ifndef XGBOOST_UTILS_CONFIG_H_
#define XGBOOST_UTILS_CONFIG_H_
#include <cstdio>
#include <cstring>
#include <string>
@ -14,26 +16,26 @@
namespace xgboost {
namespace utils {
/*!
/*!
* \brief base implementation of config reader
*/
class ConfigReaderBase {
public:
/*!
/*!
* \brief get current name, called after Next returns true
* \return current parameter name
* \return current parameter name
*/
inline const char *name(void) const {
return s_name.c_str();
}
/*!
/*!
* \brief get current value, called after Next returns true
* \return current parameter value
* \return current parameter value
*/
inline const char *val(void) const {
return s_val.c_str();
}
/*!
/*!
* \brief move iterator to next position
* \return true if there is value in next position
*/
@ -55,7 +57,7 @@ class ConfigReaderBase {
protected:
/*!
* \brief to be implemented by subclass,
* get next token, return EOF if end of file
* get next token, return EOF if end of file
*/
virtual char GetChar(void) = 0;
/*! \brief to be implemented by child, check if end of stream */
@ -144,9 +146,9 @@ class ConfigReaderBase {
*/
class ConfigStreamReader: public ConfigReaderBase {
public:
/*!
* \brief constructor
* \param istream input stream
/*!
* \brief constructor
* \param istream input stream
*/
explicit ConfigStreamReader(std::istream &fin) : fin(fin) {}
@ -163,13 +165,13 @@ class ConfigStreamReader: public ConfigReaderBase {
std::istream &fin;
};
/*!
/*!
* \brief an iterator that iterates over a configure file and gets the configures
*/
class ConfigIterator: public ConfigStreamReader {
public:
/*!
* \brief constructor
/*!
* \brief constructor
* \param fname name of configure file
*/
explicit ConfigIterator(const char *fname) : ConfigStreamReader(fi) {

20
src/utils/group_data.h
View File

@ -1,6 +1,5 @@
#ifndef XGBOOST_UTILS_GROUP_DATA_H_
#define XGBOOST_UTILS_GROUP_DATA_H_
/*!
* Copyright 2014 by Contributors
* \file group_data.h
* \brief this file defines utils to group data by integer keys
* Input: given input sequence (key,value), (k1,v1), (k2,v2)
@ -12,6 +11,11 @@
* The major algorithm is a two pass linear scan algorithm that requires two pass scan over the data
* \author Tianqi Chen
*/
#ifndef XGBOOST_UTILS_GROUP_DATA_H_
#define XGBOOST_UTILS_GROUP_DATA_H_
#include <vector>
namespace xgboost {
namespace utils {
/*!
@ -32,10 +36,10 @@ struct ParallelGroupBuilder {
std::vector< std::vector<SizeType> > *p_thread_rptr)
: rptr(*p_rptr), data(*p_data), thread_rptr(*p_thread_rptr) {
}
public:
/*!
* \brief step 1: initialize the helper, with hint of number keys
* \brief step 1: initialize the helper, with hint of number keys
* and thread used in the construction
* \param nkeys number of keys in the matrix, can be smaller than expected
* \param nthread number of thread that will be used in construction
@ -56,7 +60,7 @@ struct ParallelGroupBuilder {
inline void AddBudget(size_t key, int threadid, SizeType nelem = 1) {
std::vector<SizeType> &trptr = thread_rptr[threadid];
if (trptr.size() < key + 1) {
trptr.resize(key + 1, 0);
trptr.resize(key + 1, 0);
}
trptr[key] += nelem;
}
@ -84,13 +88,13 @@ struct ParallelGroupBuilder {
data.resize(start);
}
/*!
* \brief step 4: add data to the allocated space,
* \brief step 4: add data to the allocated space,
* the calls to this function should be exactly match previous call to AddBudget
*
* \param key the key of
* \param key the key of
* \param threadid the id of thread that calls this function
*/
inline void Push(size_t key, ValueType value, int threadid) {
inline void Push(size_t key, ValueType value, int threadid) {
SizeType &rp = thread_rptr[threadid][key];
data[rp++] = value;
}

26
src/utils/io.h
View File

@ -1,16 +1,19 @@
#ifndef XGBOOST_UTILS_IO_H
#define XGBOOST_UTILS_IO_H
/*!
* Copyright 2014 by Contributors
* \file io.h
* \brief general stream interface for serialization, I/O
* \author Tianqi Chen
*/
#ifndef XGBOOST_UTILS_IO_H_
#define XGBOOST_UTILS_IO_H_
#include <cstdio>
#include <vector>
#include <string>
#include <cstring>
#include "./utils.h"
#include "../sync/sync.h"
/*!
* \file io.h
* \brief general stream interface for serialization, I/O
* \author Tianqi Chen
*/
namespace xgboost {
namespace utils {
// reuse the definitions of streams
@ -23,7 +26,7 @@ typedef rabit::utils::MemoryBufferStream MemoryBufferStream;
class FileStream : public ISeekStream {
public:
explicit FileStream(std::FILE *fp) : fp(fp) {}
explicit FileStream(void) {
FileStream(void) {
this->fp = NULL;
}
virtual size_t Read(void *ptr, size_t size) {
@ -33,7 +36,7 @@ class FileStream : public ISeekStream {
std::fwrite(ptr, size, 1, fp);
}
virtual void Seek(size_t pos) {
std::fseek(fp, static_cast<long>(pos), SEEK_SET);
std::fseek(fp, static_cast<long>(pos), SEEK_SET); // NOLINT(*)
}
virtual size_t Tell(void) {
return std::ftell(fp);
@ -42,7 +45,7 @@ class FileStream : public ISeekStream {
return std::feof(fp) != 0;
}
inline void Close(void) {
if (fp != NULL){
if (fp != NULL) {
std::fclose(fp); fp = NULL;
}
}
@ -52,6 +55,5 @@ class FileStream : public ISeekStream {
};
} // namespace utils
} // namespace xgboost
#include "./base64-inl.h"
#endif
#endif // XGBOOST_UTILS_IO_H_

195
src/utils/thread.h
View File

@ -1,16 +1,17 @@
#ifndef XGBOOST_UTILS_THREAD_H
#define XGBOOST_UTILS_THREAD_H
/*!
* Copyright by Contributors
* \file thread.h
* \brief this header include the minimum necessary resource for multi-threading
* \brief this header include the minimum necessary resource
* for multi-threading that can be compiled in windows, linux, mac
* \author Tianqi Chen
* Acknowledgement: this file is adapted from SVDFeature project, by same author.
* The MAC support part of this code is provided by Artemy Kolchinsky
*/
#ifndef XGBOOST_UTILS_THREAD_H_ // NOLINT(*)
#define XGBOOST_UTILS_THREAD_H_ // NOLINT(*)
#ifdef _MSC_VER
#include "utils.h"
#include <windows.h>
#include <process.h>
#include "../xgboost/utils.h"
namespace xgboost {
namespace utils {
/*! \brief simple semaphore used for synchronization */
@ -18,29 +19,80 @@ class Semaphore {
public :
inline void Init(int init_val) {
sem = CreateSemaphore(NULL, init_val, 10, NULL);
utils::Assert(sem != NULL, "create Semaphore error");
utils::Check(sem != NULL, "create Semaphore error");
}
inline void Destroy(void) {
CloseHandle(sem);
}
inline void Wait(void) {
utils::Assert(WaitForSingleObject(sem, INFINITE) == WAIT_OBJECT_0, "WaitForSingleObject error");
utils::Check(WaitForSingleObject(sem, INFINITE) == WAIT_OBJECT_0, "WaitForSingleObject error");
}
inline void Post(void) {
utils::Assert(ReleaseSemaphore(sem, 1, NULL) != 0, "ReleaseSemaphore error");
utils::Check(ReleaseSemaphore(sem, 1, NULL) != 0, "ReleaseSemaphore error");
}
private:
HANDLE sem;
};
/*! \brief mutex under windows */
class Mutex {
public:
inline void Init(void) {
utils::Check(InitializeCriticalSectionAndSpinCount(&mutex, 0x00000400) != 0,
"Mutex::Init fail");
}
inline void Lock(void) {
EnterCriticalSection(&mutex);
}
inline void Unlock(void) {
LeaveCriticalSection(&mutex);
}
inline void Destroy(void) {
DeleteCriticalSection(&mutex);
}
private:
friend class ConditionVariable;
CRITICAL_SECTION mutex;
};
// conditional variable that uses pthread
class ConditionVariable {
public:
// initialize conditional variable
inline void Init(void) {
InitializeConditionVariable(&cond);
}
// destroy the thread
inline void Destroy(void) {
// DeleteConditionVariable(&cond);
}
// wait on the conditional variable
inline void Wait(Mutex *mutex) {
utils::Check(SleepConditionVariableCS(&cond, &(mutex->mutex), INFINITE) != 0,
"ConditionVariable:Wait fail");
}
inline void Broadcast(void) {
WakeAllConditionVariable(&cond);
}
inline void Signal(void) {
WakeConditionVariable(&cond);
}
private:
CONDITION_VARIABLE cond;
};
/*! \brief simple thread that wraps windows thread */
class Thread {
private:
HANDLE thread_handle;
unsigned thread_id;
unsigned thread_id;
public:
inline void Start(unsigned int __stdcall entry(void*), void *param) {
inline void Start(unsigned int __stdcall entry(void*p), void *param) {
thread_handle = (HANDLE)_beginthreadex(NULL, 0, entry, param, 0, &thread_id);
}
}
inline int Join(void) {
WaitForSingleObject(thread_handle, INFINITE);
return 0;
@ -54,39 +106,41 @@ inline void ThreadExit(void *status) {
} // namespace utils
} // namespace xgboost
#else
// thread interface using g++
extern "C" {
// thread interface using g++
#include <semaphore.h>
#include <pthread.h>
}
#include <errno.h>
namespace xgboost {
namespace utils {
/*!\brief semaphore class */
class Semaphore {
#ifdef __APPLE__
private:
sem_t* semPtr;
char sema_name[20];
char sema_name[20];
private:
inline void GenRandomString(char *s, const int len) {
static const char alphanum[] = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ" ;
static const char alphanum[] = "0123456789ABCDEFGHIJKLMNOPQRSTUVWXYZ";
for (int i = 0; i < len; ++i) {
s[i] = alphanum[rand() % (sizeof(alphanum) - 1)];
}
s[len] = 0;
}
public:
inline void Init(int init_val) {
sema_name[0]='/';
sema_name[1]='s';
sema_name[2]='e';
sema_name[3]='/';
sema_name[0] = '/';
sema_name[1] = 's';
sema_name[2] = 'e';
sema_name[3] = '/';
GenRandomString(&sema_name[4], 16);
if((semPtr = sem_open(sema_name, O_CREAT, 0644, init_val)) == SEM_FAILED) {
if ((semPtr = sem_open(sema_name, O_CREAT, 0644, init_val)) == SEM_FAILED) {
perror("sem_open");
exit(1);
}
utils::Assert(semPtr != NULL, "create Semaphore error");
utils::Check(semPtr != NULL, "create Semaphore error");
}
inline void Destroy(void) {
if (sem_close(semPtr) == -1) {
@ -103,53 +157,93 @@ class Semaphore {
}
inline void Post(void) {
sem_post(semPtr);
}
}
#else
private:
sem_t sem;
public:
inline void Init(int init_val) {
sem_init(&sem, 0, init_val);
if (sem_init(&sem, 0, init_val) != 0) {
utils::Error("Semaphore.Init:%s", strerror(errno));
}
}
inline void Destroy(void) {
sem_destroy(&sem);
if (sem_destroy(&sem) != 0) {
utils::Error("Semaphore.Destroy:%s", strerror(errno));
}
}
inline void Wait(void) {
sem_wait(&sem);
if (sem_wait(&sem) != 0) {
utils::Error("Semaphore.Wait:%s", strerror(errno));
}
}
inline void Post(void) {
sem_post(&sem);
if (sem_post(&sem) != 0) {
utils::Error("Semaphore.Post:%s", strerror(errno));
}
}
#endif
#endif
};
// helper for c thread
// used to strictly call c++ function from pthread
struct ThreadContext {
void *(*entry)(void*);
void *param;
};
extern "C" {
inline void *RunThreadContext(void *ctx_) {
ThreadContext *ctx = reinterpret_cast<ThreadContext*>(ctx_);
void *ret = (*ctx->entry)(ctx->param);
delete ctx;
return ret;
// mutex that works with pthread
class Mutex {
public:
inline void Init(void) {
pthread_mutex_init(&mutex, NULL);
}
}
inline void Lock(void) {
pthread_mutex_lock(&mutex);
}
inline void Unlock(void) {
pthread_mutex_unlock(&mutex);
}
inline void Destroy(void) {
pthread_mutex_destroy(&mutex);
}
private:
friend class ConditionVariable;
pthread_mutex_t mutex;
};
// conditional variable that uses pthread
class ConditionVariable {
public:
// initialize conditional variable
inline void Init(void) {
pthread_cond_init(&cond, NULL);
}
// destroy the thread
inline void Destroy(void) {
pthread_cond_destroy(&cond);
}
// wait on the conditional variable
inline void Wait(Mutex *mutex) {
pthread_cond_wait(&cond, &(mutex->mutex));
}
inline void Broadcast(void) {
pthread_cond_broadcast(&cond);
}
inline void Signal(void) {
pthread_cond_signal(&cond);
}
private:
pthread_cond_t cond;
};
/*!\brief simple thread class */
class Thread {
private:
pthread_t thread;
pthread_t thread;
public :
inline void Start(void *entry(void*), void *param) {
inline void Start(void * entry(void*), void *param) { // NOLINT(*)
pthread_attr_t attr;
pthread_attr_init(&attr);
pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_JOINABLE);
ThreadContext *ctx = new ThreadContext();
ctx->entry = entry; ctx->param = param;
pthread_create(&thread, &attr, RunThreadContext, ctx);
pthread_create(&thread, &attr, entry, param);
}
inline int Join(void) {
void *status;
@ -159,9 +253,8 @@ class Thread {
inline void ThreadExit(void *status) {
pthread_exit(status);
}
} // namespace utils
} // namespace xgboost
#define XGBOOST_THREAD_PREFIX void *
#endif
#endif
#endif // Linux
#endif // XGBOOST_UTILS_THREAD_H_ NOLINT(*)