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

Support dataframe data format in native XGBoost. (#9828)

Browse Source 
- Implement a columnar adapter.
- Refactor Python pandas handling code to avoid converting into a single numpy array.
- Add support in R for transforming columns.
- Support R data.frame and factor type.
This commit is contained in:
Jiaming Yuan
2023-12-12 09:56:31 +08:00
committed by GitHub
parent b3700bbb3f
commit faf0f2df10
21 changed files with 718 additions and 221 deletions
Download Patch File Download Diff File Expand all files Collapse all files

76
src/c_api/c_api.cc
View File

@@ -361,49 +361,57 @@ XGB_DLL int XGProxyDMatrixCreate(DMatrixHandle *out) {
API_END();
}
XGB_DLL int
XGProxyDMatrixSetDataCudaArrayInterface(DMatrixHandle handle,
char const *c_interface_str) {
XGB_DLL int XGProxyDMatrixSetDataCudaArrayInterface(DMatrixHandle handle,
char const *c_interface_str) {
API_BEGIN();
CHECK_HANDLE();
xgboost_CHECK_C_ARG_PTR(c_interface_str);
auto p_m = static_cast<std::shared_ptr<xgboost::DMatrix> *>(handle);
CHECK(p_m);
auto m = static_cast<xgboost::data::DMatrixProxy*>(p_m->get());
auto m = static_cast<xgboost::data::DMatrixProxy *>(p_m->get());
CHECK(m) << "Current DMatrix type does not support set data.";
m->SetCUDAArray(c_interface_str);
API_END();
}
XGB_DLL int XGProxyDMatrixSetDataCudaColumnar(DMatrixHandle handle,
char const *c_interface_str) {
XGB_DLL int XGProxyDMatrixSetDataCudaColumnar(DMatrixHandle handle, char const *c_interface_str) {
API_BEGIN();
CHECK_HANDLE();
xgboost_CHECK_C_ARG_PTR(c_interface_str);
auto p_m = static_cast<std::shared_ptr<xgboost::DMatrix> *>(handle);
CHECK(p_m);
auto m = static_cast<xgboost::data::DMatrixProxy*>(p_m->get());
auto m = static_cast<xgboost::data::DMatrixProxy *>(p_m->get());
CHECK(m) << "Current DMatrix type does not support set data.";
m->SetCUDAArray(c_interface_str);
API_END();
}
XGB_DLL int XGProxyDMatrixSetDataDense(DMatrixHandle handle,
char const *c_interface_str) {
XGB_DLL int XGProxyDMatrixSetDataColumnar(DMatrixHandle handle, char const *c_interface_str) {
API_BEGIN();
CHECK_HANDLE();
xgboost_CHECK_C_ARG_PTR(c_interface_str);
auto p_m = static_cast<std::shared_ptr<xgboost::DMatrix> *>(handle);
CHECK(p_m);
auto m = static_cast<xgboost::data::DMatrixProxy*>(p_m->get());
auto m = static_cast<xgboost::data::DMatrixProxy *>(p_m->get());
CHECK(m) << "Current DMatrix type does not support set data.";
m->SetColumnarData(c_interface_str);
API_END();
}
XGB_DLL int XGProxyDMatrixSetDataDense(DMatrixHandle handle, char const *c_interface_str) {
API_BEGIN();
CHECK_HANDLE();
xgboost_CHECK_C_ARG_PTR(c_interface_str);
auto p_m = static_cast<std::shared_ptr<xgboost::DMatrix> *>(handle);
CHECK(p_m);
auto m = static_cast<xgboost::data::DMatrixProxy *>(p_m->get());
CHECK(m) << "Current DMatrix type does not support set data.";
m->SetArrayData(c_interface_str);
API_END();
}
XGB_DLL int XGProxyDMatrixSetDataCSR(DMatrixHandle handle, char const *indptr,
char const *indices, char const *data,
xgboost::bst_ulong ncol) {
XGB_DLL int XGProxyDMatrixSetDataCSR(DMatrixHandle handle, char const *indptr, char const *indices,
char const *data, xgboost::bst_ulong ncol) {
API_BEGIN();
CHECK_HANDLE();
xgboost_CHECK_C_ARG_PTR(indptr);
@@ -411,7 +419,7 @@ XGB_DLL int XGProxyDMatrixSetDataCSR(DMatrixHandle handle, char const *indptr,
xgboost_CHECK_C_ARG_PTR(data);
auto p_m = static_cast<std::shared_ptr<xgboost::DMatrix> *>(handle);
CHECK(p_m);
auto m = static_cast<xgboost::data::DMatrixProxy*>(p_m->get());
auto m = static_cast<xgboost::data::DMatrixProxy *>(p_m->get());
CHECK(m) << "Current DMatrix type does not support set data.";
m->SetCSRData(indptr, indices, data, ncol, true);
API_END();
@@ -429,6 +437,25 @@ XGB_DLL int XGDMatrixCreateFromCSREx(const size_t *indptr, const unsigned *indic
API_END();
}
XGB_DLL int XGDMatrixCreateFromColumnar(char const *data, char const *c_json_config,
DMatrixHandle *out) {
API_BEGIN();
xgboost_CHECK_C_ARG_PTR(c_json_config);
xgboost_CHECK_C_ARG_PTR(data);
auto config = Json::Load(c_json_config);
float missing = GetMissing(config);
auto n_threads = OptionalArg<Integer, std::int64_t>(config, "nthread", 0);
auto data_split_mode =
static_cast<DataSplitMode>(OptionalArg<Integer, int64_t>(config, "data_split_mode", 0));
data::ColumnarAdapter adapter{data};
*out = new std::shared_ptr<DMatrix>(
DMatrix::Create(&adapter, missing, n_threads, "", data_split_mode));
API_END();
}
XGB_DLL int XGDMatrixCreateFromCSR(char const *indptr, char const *indices, char const *data,
xgboost::bst_ulong ncol, char const *c_json_config,
DMatrixHandle *out) {
@@ -1196,6 +1223,27 @@ XGB_DLL int XGBoosterPredictFromDense(BoosterHandle handle, char const *array_in
API_END();
}
XGB_DLL int XGBoosterPredictFromColumnar(BoosterHandle handle, char const *array_interface,
char const *c_json_config, DMatrixHandle m,
xgboost::bst_ulong const **out_shape,
xgboost::bst_ulong *out_dim, const float **out_result) {
API_BEGIN();
CHECK_HANDLE();
std::shared_ptr<DMatrix> p_m{nullptr};
if (!m) {
p_m.reset(new data::DMatrixProxy);
} else {
p_m = *static_cast<std::shared_ptr<DMatrix> *>(m);
}
auto proxy = dynamic_cast<data::DMatrixProxy *>(p_m.get());
CHECK(proxy) << "Invalid input type for inplace predict.";
xgboost_CHECK_C_ARG_PTR(array_interface);
proxy->SetColumnarData(array_interface);
auto *learner = static_cast<xgboost::Learner *>(handle);
InplacePredictImpl(p_m, c_json_config, learner, out_shape, out_dim, out_result);
API_END();
}
XGB_DLL int XGBoosterPredictFromCSR(BoosterHandle handle, char const *indptr, char const *indices,
char const *data, xgboost::bst_ulong cols,
char const *c_json_config, DMatrixHandle m,