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

Thread safe, inplace prediction. (#5389)

Browse Source 
Normal prediction with DMatrix is now thread safe with locks.  Added inplace prediction is lock free thread safe.

When data is on device (cupy, cudf), the returned data is also on device.

* Implementation for numpy, csr, cudf and cupy.

* Implementation for dask.

* Remove sync in simple dmatrix.
This commit is contained in:
Jiaming Yuan
2020-03-30 15:35:28 +08:00
committed by GitHub
parent 7f980e9f83
commit 6601a641d7
25 changed files with 1217 additions and 167 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
tests/cpp/helpers.h
View File

@@ -181,6 +181,9 @@ class RandomDataGenerator {
int32_t device_;
int32_t seed_;
Json ArrayInterfaceImpl(HostDeviceVector<float> *storage, size_t rows,
size_t cols) const;
public:
RandomDataGenerator(bst_row_t rows, size_t cols, float sparsity)
: rows_{rows}, cols_{cols}, sparsity_{sparsity}, lower_{0.0f}, upper_{1.0f},
@@ -204,7 +207,9 @@ class RandomDataGenerator {
}
void GenerateDense(HostDeviceVector<float>* out) const;
void GenerateArrayInterface(HostDeviceVector<float>* storage, std::string* out) const;
std::string GenerateArrayInterface(HostDeviceVector<float>* storage) const;
std::string GenerateColumnarArrayInterface(
std::vector<HostDeviceVector<float>> *data) const;
void GenerateCSR(HostDeviceVector<float>* value, HostDeviceVector<bst_row_t>* row_ptr,
HostDeviceVector<bst_feature_t>* columns) const;