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

[BP] Fix early stopping with linear model. (#7554) (#7562)

Browse Source
This commit is contained in:
Jiaming Yuan 2022-01-14 00:22:08 +08:00 committed by GitHub
parent 35dac8af1d
commit 87ddcf308e
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 34 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
R-package/tests/testthat/test_glm.R
View File

@ -46,3 +46,31 @@ test_that("gblinear works", {
expect_equal(dim(h), c(n, ncol(dtrain) + 1)) expect_equal(dim(h), c(n, ncol(dtrain) + 1))
expect_s4_class(h, "dgCMatrix") expect_s4_class(h, "dgCMatrix")
}) })
test_that("gblinear early stopping works", {
data(agaricus.train, package = 'xgboost')
data(agaricus.test, package = 'xgboost')
dtrain <- xgb.DMatrix(agaricus.train$data, label = agaricus.train$label)
dtest <- xgb.DMatrix(agaricus.test$data, label = agaricus.test$label)
param <- list(
objective = "binary:logistic", eval_metric = "error", booster = "gblinear",
nthread = 2, eta = 0.8, alpha = 0.0001, lambda = 0.0001,
updater = "coord_descent"
)
es_round <- 1
n <- 10
booster <- xgb.train(
param, dtrain, n, list(eval = dtest, train = dtrain), early_stopping_rounds = es_round
)
expect_equal(booster$best_iteration, 5)
predt_es <- predict(booster, dtrain)
n <- booster$best_iteration + es_round
booster <- xgb.train(
param, dtrain, n, list(eval = dtest, train = dtrain), early_stopping_rounds = es_round
)
predt <- predict(booster, dtrain)
expect_equal(predt_es, predt)
})

13
src/gbm/gblinear.cc
View File

@ -62,9 +62,8 @@ struct GBLinearTrainParam : public XGBoostParameter<GBLinearTrainParam> {
} }
}; };
void LinearCheckLayer(unsigned layer_begin, unsigned layer_end) { void LinearCheckLayer(unsigned layer_begin) {
CHECK_EQ(layer_begin, 0) << "Linear booster does not support prediction range."; CHECK_EQ(layer_begin, 0) << "Linear booster does not support prediction range.";
CHECK_EQ(layer_end, 0) << "Linear booster does not support prediction range.";
} }
/*! /*!
@ -152,7 +151,7 @@ class GBLinear : public GradientBooster {
void PredictBatch(DMatrix *p_fmat, PredictionCacheEntry *predts, void PredictBatch(DMatrix *p_fmat, PredictionCacheEntry *predts,
bool training, unsigned layer_begin, unsigned layer_end) override { bool training, unsigned layer_begin, unsigned layer_end) override {
monitor_.Start("PredictBatch"); monitor_.Start("PredictBatch");
LinearCheckLayer(layer_begin, layer_end); LinearCheckLayer(layer_begin);
auto* out_preds = &predts->predictions; auto* out_preds = &predts->predictions;
this->PredictBatchInternal(p_fmat, &out_preds->HostVector()); this->PredictBatchInternal(p_fmat, &out_preds->HostVector());
monitor_.Stop("PredictBatch"); monitor_.Stop("PredictBatch");
@ -161,7 +160,7 @@ class GBLinear : public GradientBooster {
void PredictInstance(const SparsePage::Inst &inst, void PredictInstance(const SparsePage::Inst &inst,
std::vector<bst_float> *out_preds, std::vector<bst_float> *out_preds,
unsigned layer_begin, unsigned layer_end) override { unsigned layer_begin, unsigned layer_end) override {
LinearCheckLayer(layer_begin, layer_end); LinearCheckLayer(layer_begin);
const int ngroup = model_.learner_model_param->num_output_group; const int ngroup = model_.learner_model_param->num_output_group;
for (int gid = 0; gid < ngroup; ++gid) { for (int gid = 0; gid < ngroup; ++gid) {
this->Pred(inst, dmlc::BeginPtr(*out_preds), gid, this->Pred(inst, dmlc::BeginPtr(*out_preds), gid,
@ -177,8 +176,8 @@ class GBLinear : public GradientBooster {
HostDeviceVector<bst_float>* out_contribs, HostDeviceVector<bst_float>* out_contribs,
unsigned layer_begin, unsigned layer_end, bool, int, unsigned) override { unsigned layer_begin, unsigned layer_end, bool, int, unsigned) override {
model_.LazyInitModel(); model_.LazyInitModel();
LinearCheckLayer(layer_begin, layer_end); LinearCheckLayer(layer_begin);
const auto& base_margin = p_fmat->Info().base_margin_.ConstHostVector(); const auto &base_margin = p_fmat->Info().base_margin_.ConstHostVector();
const int ngroup = model_.learner_model_param->num_output_group; const int ngroup = model_.learner_model_param->num_output_group;
const size_t ncolumns = model_.learner_model_param->num_feature + 1; const size_t ncolumns = model_.learner_model_param->num_feature + 1;
// allocate space for (#features + bias) times #groups times #rows // allocate space for (#features + bias) times #groups times #rows
@ -214,7 +213,7 @@ class GBLinear : public GradientBooster {
void PredictInteractionContributions(DMatrix* p_fmat, void PredictInteractionContributions(DMatrix* p_fmat,
HostDeviceVector<bst_float>* out_contribs, HostDeviceVector<bst_float>* out_contribs,
unsigned layer_begin, unsigned layer_end, bool) override { unsigned layer_begin, unsigned layer_end, bool) override {
LinearCheckLayer(layer_begin, layer_end); LinearCheckLayer(layer_begin);
std::vector<bst_float>& contribs = out_contribs->HostVector(); std::vector<bst_float>& contribs = out_contribs->HostVector();
// linear models have no interaction effects // linear models have no interaction effects