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

Unify logging facilities. (#3982)

Browse Source 
* Unify logging facilities.

* Enhance `ConsoleLogger` to handle different verbosity.
* Override macros from `dmlc`.
* Don't use specialized gamma when building with GPU.
* Remove verbosity cache in monitor.
* Test monitor.
* Deprecate `silent`.
* Fix doc and messages.
* Fix python test.
* Fix silent tests.
This commit is contained in:
Jiaming Yuan
2018-12-14 19:29:58 +08:00
committed by GitHub
parent fd722d60cd
commit e0a279114e
28 changed files with 368 additions and 171 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
src/tree/param.h
View File

@@ -34,8 +34,7 @@ struct TrainParam : public dmlc::Parameter<TrainParam> {
// growing policy
enum TreeGrowPolicy { kDepthWise = 0, kLossGuide = 1 };
int grow_policy;
// flag to print out detailed breakdown of runtime
int debug_verbose;
//----- the rest parameters are less important ----
// minimum amount of hessian(weight) allowed in a child
float min_child_weight;
@@ -67,8 +66,6 @@ struct TrainParam : public dmlc::Parameter<TrainParam> {
int parallel_option;
// option to open cacheline optimization
bool cache_opt;
// whether to not print info during training.
bool silent;
// whether refresh updater needs to update the leaf values
bool refresh_leaf;
// auxiliary data structure
@@ -107,10 +104,6 @@ struct TrainParam : public dmlc::Parameter<TrainParam> {
.set_default(0.0f)
.describe(
"Minimum loss reduction required to make a further partition.");
DMLC_DECLARE_FIELD(debug_verbose)
.set_lower_bound(0)
.set_default(0)
.describe("flag to print out detailed breakdown of runtime");
DMLC_DECLARE_FIELD(max_depth)
.set_lower_bound(0)
.set_default(6)
@@ -186,9 +179,6 @@ struct TrainParam : public dmlc::Parameter<TrainParam> {
DMLC_DECLARE_FIELD(cache_opt)
.set_default(true)
.describe("EXP Param: Cache aware optimization.");
DMLC_DECLARE_FIELD(silent)
.set_default(false)
.describe("Do not print information during trainig.");
DMLC_DECLARE_FIELD(refresh_leaf)
.set_default(true)
.describe("Whether the refresh updater needs to update leaf values.");

2
src/tree/updater_gpu.cu
View File

@@ -625,7 +625,7 @@ class GPUMaker : public TreeUpdater {
void allocateAllData(int offsetSize) {
int tmpBuffSize = ScanTempBufferSize(nVals);
ba.Allocate(param.gpu_id, param.silent, &vals, nVals,
ba.Allocate(param.gpu_id, &vals, nVals,
&vals_cached, nVals, &instIds, nVals, &instIds_cached, nVals,
&colOffsets, offsetSize, &gradsInst, nRows, &nodeAssigns, nVals,
&nodeLocations, nVals, &nodes, maxNodes, &nodeAssignsPerInst,

10
src/tree/updater_gpu_hist.cu
View File

@@ -803,7 +803,7 @@ inline void DeviceShard<GradientSumT>::InitCompressedData(
int max_nodes =
param.max_leaves > 0 ? param.max_leaves * 2 : MaxNodesDepth(param.max_depth);
ba.Allocate(device_id_, param.silent,
ba.Allocate(device_id_,
&gpair, n_rows,
&ridx, n_rows,
&position, n_rows,
@@ -833,7 +833,7 @@ inline void DeviceShard<GradientSumT>::InitCompressedData(
CHECK(!(param.max_leaves == 0 && param.max_depth == 0))
<< "Max leaves and max depth cannot both be unconstrained for "
"gpu_hist.";
ba.Allocate(device_id_, param.silent, &gidx_buffer, compressed_size_bytes);
ba.Allocate(device_id_, &gidx_buffer, compressed_size_bytes);
gidx_buffer.Fill(0);
int nbits = common::detail::SymbolBits(num_symbols);
@@ -931,7 +931,7 @@ class GPUHistMakerSpecialised{
qexpand_.reset(new ExpandQueue(DepthWise));
}
monitor_.Init("updater_gpu_hist", param_.debug_verbose);
monitor_.Init("updater_gpu_hist");
}
void Update(HostDeviceVector<GradientPair>* gpair, DMatrix* dmat,
@@ -966,7 +966,9 @@ class GPUHistMakerSpecialised{
device_list_[index] = device_id;
}
reducer_.Init(device_list_, param_.debug_verbose);
reducer_.Init(
device_list_,
ConsoleLogger::GlobalVerbosity() > ConsoleLogger::DefaultVerbosity());
auto batch_iter = dmat->GetRowBatches().begin();
const SparsePage& batch = *batch_iter;

8
src/tree/updater_prune.cc
View File

@@ -71,11 +71,9 @@ class TreePruner: public TreeUpdater {
npruned = this->TryPruneLeaf(tree, nid, tree.GetDepth(nid), npruned);
}
}
if (!param_.silent) {
LOG(INFO) << "tree pruning end, " << tree.param.num_roots << " roots, "
<< tree.NumExtraNodes() << " extra nodes, " << npruned
<< " pruned nodes, max_depth=" << tree.MaxDepth();
}
LOG(INFO) << "tree pruning end, " << tree.param.num_roots << " roots, "
<< tree.NumExtraNodes() << " extra nodes, " << npruned
<< " pruned nodes, max_depth=" << tree.MaxDepth();
}
private:

57
src/tree/updater_quantile_hist.cc
View File

@@ -6,6 +6,7 @@
*/
#include <dmlc/timer.h>
#include <rabit/rabit.h>
#include <xgboost/logging.h>
#include <xgboost/tree_updater.h>
#include <cmath>
@@ -60,9 +61,7 @@ void QuantileHistMaker::Update(HostDeviceVector<GradientPair> *gpair,
gmatb_.Init(gmat_, column_matrix_, param_);
}
is_gmat_initialized_ = true;
if (param_.debug_verbose > 0) {
LOG(INFO) << "Generating gmat: " << dmlc::GetTime() - tstart << " sec";
}
LOG(INFO) << "Generating gmat: " << dmlc::GetTime() - tstart << " sec";
}
// rescale learning rate according to size of trees
float lr = param_.learning_rate;
@@ -207,32 +206,34 @@ void QuantileHistMaker::Builder::Update(const GHistIndexMatrix& gmat,
pruner_->Update(gpair, p_fmat, std::vector<RegTree*>{p_tree});
if (param_.debug_verbose > 0) {
double total_time = dmlc::GetTime() - gstart;
LOG(INFO) << "\nInitData: "
<< std::fixed << std::setw(6) << std::setprecision(4) << time_init_data
<< " (" << std::fixed << std::setw(5) << std::setprecision(2)
<< time_init_data / total_time * 100 << "%)\n"
<< "InitNewNode: "
<< std::fixed << std::setw(6) << std::setprecision(4) << time_init_new_node
<< " (" << std::fixed << std::setw(5) << std::setprecision(2)
<< time_init_new_node / total_time * 100 << "%)\n"
<< "BuildHist: "
<< std::fixed << std::setw(6) << std::setprecision(4) << time_build_hist
<< " (" << std::fixed << std::setw(5) << std::setprecision(2)
<< time_build_hist / total_time * 100 << "%)\n"
<< "EvaluateSplit: "
<< std::fixed << std::setw(6) << std::setprecision(4) << time_evaluate_split
<< " (" << std::fixed << std::setw(5) << std::setprecision(2)
<< time_evaluate_split / total_time * 100 << "%)\n"
<< "ApplySplit: "
<< std::fixed << std::setw(6) << std::setprecision(4) << time_apply_split
<< " (" << std::fixed << std::setw(5) << std::setprecision(2)
<< time_apply_split / total_time * 100 << "%)\n"
<< "========================================\n"
<< "Total: "
<< std::fixed << std::setw(6) << std::setprecision(4) << total_time;
if (ConsoleLogger::GlobalVerbosity() <= ConsoleLogger::DefaultVerbosity()) {
// Don't construct the following huge stream.
return;
}
double total_time = dmlc::GetTime() - gstart;
LOG(INFO) << "\nInitData: "
<< std::fixed << std::setw(6) << std::setprecision(4) << time_init_data
<< " (" << std::fixed << std::setw(5) << std::setprecision(2)
<< time_init_data / total_time * 100 << "%)\n"
<< "InitNewNode: "
<< std::fixed << std::setw(6) << std::setprecision(4) << time_init_new_node
<< " (" << std::fixed << std::setw(5) << std::setprecision(2)
<< time_init_new_node / total_time * 100 << "%)\n"
<< "BuildHist: "
<< std::fixed << std::setw(6) << std::setprecision(4) << time_build_hist
<< " (" << std::fixed << std::setw(5) << std::setprecision(2)
<< time_build_hist / total_time * 100 << "%)\n"
<< "EvaluateSplit: "
<< std::fixed << std::setw(6) << std::setprecision(4) << time_evaluate_split
<< " (" << std::fixed << std::setw(5) << std::setprecision(2)
<< time_evaluate_split / total_time * 100 << "%)\n"
<< "ApplySplit: "
<< std::fixed << std::setw(6) << std::setprecision(4) << time_apply_split
<< " (" << std::fixed << std::setw(5) << std::setprecision(2)
<< time_apply_split / total_time * 100 << "%)\n"
<< "========================================\n"
<< "Total: "
<< std::fixed << std::setw(6) << std::setprecision(4) << total_time;
}
bool QuantileHistMaker::Builder::UpdatePredictionCache(