why java predict interface only support single thread · Issue #3058 · dmlc/xgboost

相关文章推荐

讲道义的脸盆 · 解析或序列化 XML - ...· 2 月前 ·

跑龙套的单杠 · 分享6个对象数组去重的方法开发者社区· 2 月前 ·

纯真的炒饭 · MongoDB数据库中更新与删除数据 - ...· 2 月前 ·

体贴的红金鱼 · 在 Office 加载项中使用 ...· 2 月前 ·

满身肌肉的八宝粥 · 【Day31】ChatGPT請教教我：Jes ...· 1 月前 ·

另类的地瓜 · 尹力分别会见德国奔驰集团和日本瑞穗金融集团负责人· 5 月前 ·

纯真的电脑桌 · 小小保温杯，竟能变“炸弹”！保温杯千万不要这 ...· 1 年前 ·

神勇威武的小蝌蚪 · visio用例图箭头怎么画_Visio画图( ...· 1 年前 ·

热心肠的冲锋衣 · 中华书局点校本二十四史及清史稿-宋史（四） ...· 1 年前 ·

睡不着的毛衣 · google drive proxy ...· 2 年前 ·

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

I want to use java predict interface(float[][] predicts = booster.predict(testMat);) as online service. But I found it does not support multithreading. Because synchronized used in predict process. Why lock the predict. When I remove key word "synchronized" in function "private synchronized float[][] predict(DMatrix data, boolean outputMargin, int treeLimit, boolean predLeaf, boolean predContribs) throws XGBoostError", the core dump happend. How can i fix this bug. @tqchen @CodingCat @fromradio

@Craigacp @tqchen @phunterlau . i have already fix two unsafe thread call in c++ source code. Because mutithread write share variable at same time.
when i call predict function, the internal function PredLoopSpecalize will be used. there are two part of code which could cause unsafe thread(note that other function maybe have unsafe thread problem, but i just fix my call).

thread_temp useing in code as follow is share variable, when multithread write into thread_temp, it cause core dump. i change the thread_temp to local variable.
inline void PredLoopSpecalize(DMatrix* p_fmat,
std::vector<bst_float>* out_preds,
const gbm::GBTreeModel& model, int num_group,
unsigned tree_begin, unsigned tree_end) {
const MetaInfo& info = p_fmat->info();

const int nthread = omp_get_max_threads();
InitThreadTemp(nthread, model.param.num_feature);
std::vector<bst_float>& preds = out_preds;
CHECK_EQ(model.param.size_leaf_vector, 0)
<< "size_leaf_vector is enforced to 0 so far";
CHECK_EQ(preds.size(), p_fmat->info().num_row * num_group);
// start collecting the prediction
dmlc::DataIter iter = p_fmat->RowIterator();
iter->BeforeFirst();
int b = 0;
while (iter->BatchNext()) {
const RowBatch& batch = iter->Value();
// parallel over local batch
const int K = 8;
const bst_omp_uint nsize = static_cast<bst_omp_uint>(batch.size);
const bst_omp_uint rest = nsize % K;
#pragma omp parallel for schedule(static)
for (bst_omp_uint i = 0; i < nsize - rest; i += K) {
const int tid = omp_get_thread_num();
RegTree::FVec& feats = thread_temp [tid];

  int64_t ridx[K];
  RowBatch::Inst inst[K];
  for (int k = 0; k < K; ++k) {
    ridx[k] = static_cast<int64_t>(batch.base_rowid + i + k);
  for (int k = 0; k < K; ++k) {
    inst[k] = batch[i + k];
  for (int k = 0; k < K; ++k) {
    for (int gid = 0; gid < num_group; ++gid) {
      const size_t offset = ridx[k] * num_group + gid;
      preds[offset] += this->PredValue(
          inst[k], model.trees, model.tree_info, gid,
          info.GetRoot(ridx[k]), &feats, tree_begin, tree_end);
for (bst_omp_uint i = nsize - rest; i < nsize; ++i) {
  RegTree::FVec& feats = **thread_temp**[0];
  const int64_t ridx = static_cast<int64_t>(batch.base_rowid + i);
  const RowBatch::Inst inst = batch[i];
  for (int gid = 0; gid < num_group; ++gid) {
    const size_t offset = ridx * num_group + gid;
    preds[offset] +=
        this->PredValue(inst, model.trees, model.tree_info, gid,
                        info.GetRoot(ridx), &feats, tree_begin, tree_end);
  break;
variable at_first_ used in BeforeFirst and Next function, it can be changed during multithread call, when its  value changed, Next() will return false, then prediction stop, return defaule value.

void SimpleCSRSource::BeforeFirst() {

at_first_ = true;

bool SimpleCSRSource::Next() {

//LOG(CONSOLE)<<"SimpleCSRSource::Next()"<<"\n";

if (!at_first_) return false;

at_first_ = false;

batch_.size = row_ptr_.size() - 1;

batch_.base_rowid = 0;

batch_.ind_ptr = dmlc::BeginPtr(row_ptr_);

batch_.data_ptr = dmlc::BeginPtr(row_data_);

return true;

          
I face the same problem. hi @superclocks, how do you solve it? I did not get your point.

just use batch input, or?
          I look at the java source code, it is still synchronized. Should I assume it is not thread-safe in C++ implementation in the latest release(0.72)?

Thanks,

James
private synchronized float[][] predict(DMatrix data, boolean outputMargin, int treeLimit, boolean predLeaf, boolean predContribs) throws XGBoostError.
          @superclocks I follow your blog, but get error:
在成员函数‘virtual bool xgboost::data::SimpleCSRSource::BatchNext()’中: 
错误：‘batch_’在此作用域中尚未声明 
错误：‘row_ptr_’在此作用域中尚未声明 
错误：‘row_data_’在此作用域中尚未声明
          @superclocks After based on tag 0.72 code, the error is:
src/c_api/c_api.cc: 在成员函数‘void xgboost::Booster::LazyInit()’中:
src/c_api/c_api.cc:55:27: 错误：‘lock_’在此作用域中尚未声明
       pthread_mutex_lock(&lock_);