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

[Breaking] Accept multi-dim meta info. (#7405)

Browse Source 
This PR changes base_margin into a 3-dim array, with one of them being reserved for multi-target classification. Also, a breaking change is made for binary serialization due to extra dimension along with a fix for saving the feature weights. Lastly, it unifies the prediction initialization between CPU and GPU. After this PR, the meta info setter in Python will be based on array interface.
This commit is contained in:
Jiaming Yuan
2021-11-18 23:02:54 +08:00
committed by GitHub
parent 9fb4338964
commit d33854af1b
25 changed files with 545 additions and 256 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
src/gbm/gblinear.cc
View File

@@ -178,7 +178,7 @@ class GBLinear : public GradientBooster {
unsigned layer_begin, unsigned layer_end, bool, int, unsigned) override {
model_.LazyInitModel();
LinearCheckLayer(layer_begin, layer_end);
const auto& base_margin = p_fmat->Info().base_margin_.ConstHostVector();
auto base_margin = p_fmat->Info().base_margin_.View(GenericParameter::kCpuId);
const int ngroup = model_.learner_model_param->num_output_group;
const size_t ncolumns = model_.learner_model_param->num_feature + 1;
// allocate space for (#features + bias) times #groups times #rows
@@ -203,9 +203,9 @@ class GBLinear : public GradientBooster {
p_contribs[ins.index] = ins.fvalue * model_[ins.index][gid];
}
// add base margin to BIAS
p_contribs[ncolumns - 1] = model_.Bias()[gid] +
((base_margin.size() != 0) ? base_margin[row_idx * ngroup + gid] :
learner_model_param_->base_score);
p_contribs[ncolumns - 1] =
model_.Bias()[gid] + ((base_margin.Size() != 0) ? base_margin(row_idx, gid)
: learner_model_param_->base_score);
}
});
}
@@ -270,7 +270,7 @@ class GBLinear : public GradientBooster {
monitor_.Start("PredictBatchInternal");
model_.LazyInitModel();
std::vector<bst_float> &preds = *out_preds;
const auto& base_margin = p_fmat->Info().base_margin_.ConstHostVector();
auto base_margin = p_fmat->Info().base_margin_.View(GenericParameter::kCpuId);
// start collecting the prediction
const int ngroup = model_.learner_model_param->num_output_group;
preds.resize(p_fmat->Info().num_row_ * ngroup);
@@ -280,16 +280,15 @@ class GBLinear : public GradientBooster {
// k is number of group
// parallel over local batch
const auto nsize = static_cast<omp_ulong>(batch.Size());
if (base_margin.size() != 0) {
CHECK_EQ(base_margin.size(), nsize * ngroup);
if (base_margin.Size() != 0) {
CHECK_EQ(base_margin.Size(), nsize * ngroup);
}
common::ParallelFor(nsize, [&](omp_ulong i) {
const size_t ridx = page.base_rowid + i;
// loop over output groups
for (int gid = 0; gid < ngroup; ++gid) {
bst_float margin =
(base_margin.size() != 0) ?
base_margin[ridx * ngroup + gid] : learner_model_param_->base_score;
float margin =
(base_margin.Size() != 0) ? base_margin(ridx, gid) : learner_model_param_->base_score;
this->Pred(batch[i], &preds[ridx * ngroup], gid, margin);
}
});