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

Refactor linear modelling and add new coordinate descent updater (#3103)

Browse Source 
* Refactor linear modelling and add new coordinate descent updater

* Allow unsorted column iterator

* Add prediction cacheing to gblinear
This commit is contained in:
Rory Mitchell
2018-02-17 09:17:01 +13:00
committed by GitHub
parent 9ffe8596f2
commit 10eb05a63a
23 changed files with 1252 additions and 271 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/data/sparse_page_dmatrix.h
View File

@@ -40,8 +40,8 @@ class SparsePageDMatrix : public DMatrix {
return iter;
}
bool HaveColAccess() const override {
return col_iter_.get() != nullptr;
bool HaveColAccess(bool sorted) const override {
return col_iter_.get() != nullptr && col_iter_->sorted == sorted;
}
const RowSet& buffered_rowset() const override {
@@ -67,7 +67,7 @@ class SparsePageDMatrix : public DMatrix {
void InitColAccess(const std::vector<bool>& enabled,
float subsample,
size_t max_row_perbatch) override;
size_t max_row_perbatch, bool sorted) override;
/*! \brief page size 256 MB */
static const size_t kPageSize = 256UL << 20UL;
@@ -87,6 +87,8 @@ class SparsePageDMatrix : public DMatrix {
bool Next() override;
// initialize the column iterator with the specified index set.
void Init(const std::vector<bst_uint>& index_set, bool load_all);
// If the column features are sorted
bool sorted;
private:
// the temp page.