Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

reduce the buffer when using high dimensional data in distributed mode. #2485

Merged
merged 17 commits into from
Oct 15, 2019
Merged

reduce the buffer when using high dimensional data in distributed mode. #2485

Changes from 1 commit
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
a67f686
reduce the buffer when using high dimensional data in distributed mode.
guolinke Oct 1, 2019
958ecfb
Update dataset_loader.cpp
guolinke Oct 3, 2019
3d2b82a
Merge remote-tracking branch 'origin/master' into dist-memory-reduce
guolinke Oct 3, 2019
fc3e573
refix
guolinke Oct 3, 2019
e363201
typo
guolinke Oct 3, 2019
b1598a0
fix number of bin accumulation.
guolinke Oct 4, 2019
1618260
avoid overflow
guolinke Oct 4, 2019
7e4f2ab
fix warning
guolinke Oct 4, 2019
affcc88
efficient solution.
guolinke Oct 4, 2019
3b59e76
Update dataset.h
guolinke Oct 4, 2019
96db1a6
fix bin count output
guolinke Oct 4, 2019
a73d4fd
fix warning
guolinke Oct 4, 2019
c25557c
bug in dist number of feature check
guolinke Oct 4, 2019
1847238
fix possible edge case
guolinke Oct 4, 2019
3152a3c
Update dataset.cpp
guolinke Oct 5, 2019
25ac715
possible bug fix
guolinke Oct 8, 2019
7e1e11d
fix
guolinke Oct 8, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
33 changes: 31 additions & 2 deletions src/io/dataset_loader.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -594,7 +594,22 @@ Dataset* DatasetLoader::CostructFromSampleData(double** sample_values,

const data_size_t filter_cnt = static_cast<data_size_t>(
static_cast<double>(config_.min_data_in_leaf * total_sample_size) / num_data);
if (Network::num_machines() == 1) {

bool force_findbin_in_single_machine = false;
if (Network::num_machines() > 1) {
int total_num_feature = Network::GlobalSyncUpByMin(num_col);
size_t esimate_sync_size = BinMapper::SizeForSpecificBin(config_.max_bin) * total_num_feature;
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This will still lead to overflow, need to cast the operands before assigning to a wider type.

Suggested change
size_t esimate_sync_size = BinMapper::SizeForSpecificBin(config_.max_bin) * total_num_feature;
size_t esimate_sync_size = (size_t)BinMapper::SizeForSpecificBin(config_.max_bin) * (size_t)total_num_feature;

const size_t max_buf_size = 2 << 31;
if (esimate_sync_size >= max_buf_size) {
if (config_.pre_partition) {
Log::Warning("Too many features for distributed model, it is better to pass categorical feature directly instead of sparse high dimensional feature vectors.");
} else {
force_findbin_in_single_machine = true;
}
}
}

if (Network::num_machines() == 1 || force_findbin_in_single_machine) {
// if only one machine, find bin locally
OMP_INIT_EX();
#pragma omp parallel for schedule(guided)
Expand Down Expand Up @@ -933,8 +948,22 @@ void DatasetLoader::ConstructBinMappersFromTextData(int rank, int num_machines,
const data_size_t filter_cnt = static_cast<data_size_t>(
static_cast<double>(config_.min_data_in_leaf* sample_data.size()) / dataset->num_data_);

bool force_findbin_in_single_machine = false;
if (Network::num_machines() > 1) {
int total_num_feature = Network::GlobalSyncUpByMin(dataset->num_total_features_);
size_t esimate_sync_size = BinMapper::SizeForSpecificBin(config_.max_bin) * total_num_feature;
Copy link

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Same as above, avoids overflow.

Suggested change
size_t esimate_sync_size = BinMapper::SizeForSpecificBin(config_.max_bin) * total_num_feature;
size_t esimate_sync_size = (size_t)BinMapper::SizeForSpecificBin(config_.max_bin) * (size_t)total_num_feature;

const size_t max_buf_size = 2 << 31;
if (esimate_sync_size >= max_buf_size) {
if (config_.pre_partition) {
Log::Warning("Too many features for distributed model, it is better to pass categorical feature directly instead of sparse high dimensional feature vectors.");
} else {
force_findbin_in_single_machine = true;
}
}
}

// start find bins
if (num_machines == 1) {
if (num_machines == 1 || force_findbin_in_single_machine) {
// if only one machine, find bin locally
OMP_INIT_EX();
#pragma omp parallel for schedule(guided)
Expand Down