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

Convert labels into tensor. (#7456)

Browse Source 
* Add a new ctor to tensor for `initilizer_list`.
* Change labels from host device vector to tensor.
* Rename the field from `labels_` to `labels` since it's a public member.
This commit is contained in:
Jiaming Yuan
2021-12-17 00:58:35 +08:00
committed by GitHub
parent 6f8a4633b7
commit 5b1161bb64
35 changed files with 319 additions and 258 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/objective/hinge.cu
View File

@@ -33,11 +33,11 @@ class HingeObj : public ObjFunction {
const MetaInfo &info,
int iter,
HostDeviceVector<GradientPair> *out_gpair) override {
CHECK_NE(info.labels_.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels_.Size())
CHECK_NE(info.labels.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels.Size())
<< "labels are not correctly provided"
<< "preds.size=" << preds.Size()
<< ", label.size=" << info.labels_.Size();
<< ", label.size=" << info.labels.Size();
const size_t ndata = preds.Size();
const bool is_null_weight = info.weights_.Size() == 0;
@@ -67,7 +67,7 @@ class HingeObj : public ObjFunction {
},
common::Range{0, static_cast<int64_t>(ndata)},
tparam_->gpu_id).Eval(
out_gpair, &preds, &info.labels_, &info.weights_);
out_gpair, &preds, info.labels.Data(), &info.weights_);
}
void PredTransform(HostDeviceVector<bst_float> *io_preds) const override {

10
src/objective/multiclass_obj.cu
View File

@@ -55,13 +55,13 @@ class SoftmaxMultiClassObj : public ObjFunction {
// Remove unused parameter compiler warning.
(void) iter;
if (info.labels_.Size() == 0) {
if (info.labels.Size() == 0) {
return;
}
CHECK(preds.Size() == (static_cast<size_t>(param_.num_class) * info.labels_.Size()))
CHECK(preds.Size() == (static_cast<size_t>(param_.num_class) * info.labels.Size()))
<< "SoftmaxMultiClassObj: label size and pred size does not match.\n"
<< "label.Size() * num_class: "
<< info.labels_.Size() * static_cast<size_t>(param_.num_class) << "\n"
<< info.labels.Size() * static_cast<size_t>(param_.num_class) << "\n"
<< "num_class: " << param_.num_class << "\n"
<< "preds.Size(): " << preds.Size();
@@ -70,7 +70,7 @@ class SoftmaxMultiClassObj : public ObjFunction {
auto device = tparam_->gpu_id;
out_gpair->SetDevice(device);
info.labels_.SetDevice(device);
info.labels.SetDevice(device);
info.weights_.SetDevice(device);
preds.SetDevice(device);
@@ -115,7 +115,7 @@ class SoftmaxMultiClassObj : public ObjFunction {
gpair[idx * nclass + k] = GradientPair(p * wt, h);
}
}, common::Range{0, ndata}, device, false)
.Eval(out_gpair, &info.labels_, &preds, &info.weights_, &label_correct_);
.Eval(out_gpair, info.labels.Data(), &preds, &info.weights_, &label_correct_);
std::vector<int>& label_correct_h = label_correct_.HostVector();
for (auto const flag : label_correct_h) {

22
src/objective/rank_obj.cu
View File

@@ -760,15 +760,15 @@ class LambdaRankObj : public ObjFunction {
const MetaInfo& info,
int iter,
HostDeviceVector<GradientPair>* out_gpair) override {
CHECK_EQ(preds.Size(), info.labels_.Size()) << "label size predict size not match";
CHECK_EQ(preds.Size(), info.labels.Size()) << "label size predict size not match";
// quick consistency when group is not available
std::vector<unsigned> tgptr(2, 0); tgptr[1] = static_cast<unsigned>(info.labels_.Size());
std::vector<unsigned> tgptr(2, 0); tgptr[1] = static_cast<unsigned>(info.labels.Size());
const std::vector<unsigned> &gptr = info.group_ptr_.size() == 0 ? tgptr : info.group_ptr_;
CHECK(gptr.size() != 0 && gptr.back() == info.labels_.Size())
CHECK(gptr.size() != 0 && gptr.back() == info.labels.Size())
<< "group structure not consistent with #rows" << ", "
<< "group ponter size: " << gptr.size() << ", "
<< "labels size: " << info.labels_.Size() << ", "
<< "labels size: " << info.labels.Size() << ", "
<< "group pointer back: " << (gptr.size() == 0 ? 0 : gptr.back());
#if defined(__CUDACC__)
@@ -820,7 +820,7 @@ class LambdaRankObj : public ObjFunction {
bst_float weight_normalization_factor = ComputeWeightNormalizationFactor(info, gptr);
const auto& preds_h = preds.HostVector();
const auto& labels = info.labels_.HostVector();
const auto& labels = info.labels.HostView();
std::vector<GradientPair>& gpair = out_gpair->HostVector();
const auto ngroup = static_cast<bst_omp_uint>(gptr.size() - 1);
out_gpair->Resize(preds.Size());
@@ -841,7 +841,7 @@ class LambdaRankObj : public ObjFunction {
exc.Run([&]() {
lst.clear(); pairs.clear();
for (unsigned j = gptr[k]; j < gptr[k+1]; ++j) {
lst.emplace_back(preds_h[j], labels[j], j);
lst.emplace_back(preds_h[j], labels(j), j);
gpair[j] = GradientPair(0.0f, 0.0f);
}
std::stable_sort(lst.begin(), lst.end(), ListEntry::CmpPred);
@@ -916,7 +916,7 @@ class LambdaRankObj : public ObjFunction {
// Set the device ID and copy them to the device
out_gpair->SetDevice(device);
info.labels_.SetDevice(device);
info.labels.SetDevice(device);
preds.SetDevice(device);
info.weights_.SetDevice(device);
@@ -924,19 +924,19 @@ class LambdaRankObj : public ObjFunction {
auto d_preds = preds.ConstDevicePointer();
auto d_gpair = out_gpair->DevicePointer();
auto d_labels = info.labels_.ConstDevicePointer();
auto d_labels = info.labels.View(device);
SortedLabelList slist(param_);
// Sort the labels within the groups on the device
slist.Sort(info.labels_, gptr);
slist.Sort(*info.labels.Data(), gptr);
// Initialize the gradients next
out_gpair->Fill(GradientPair(0.0f, 0.0f));
// Finally, compute the gradients
slist.ComputeGradients<LambdaWeightComputerT>
(d_preds, d_labels, info.weights_, iter, d_gpair, weight_normalization_factor);
slist.ComputeGradients<LambdaWeightComputerT>(d_preds, d_labels.Values().data(), info.weights_,
iter, d_gpair, weight_normalization_factor);
}
#endif

35
src/objective/regression_obj.cu
View File

@@ -59,9 +59,9 @@ class RegLossObj : public ObjFunction {
void GetGradient(const HostDeviceVector<bst_float>& preds,
const MetaInfo &info, int,
HostDeviceVector<GradientPair>* out_gpair) override {
CHECK_EQ(preds.Size(), info.labels_.Size())
CHECK_EQ(preds.Size(), info.labels.Size())
<< " " << "labels are not correctly provided"
<< "preds.size=" << preds.Size() << ", label.size=" << info.labels_.Size() << ", "
<< "preds.size=" << preds.Size() << ", label.size=" << info.labels.Size() << ", "
<< "Loss: " << Loss::Name();
size_t const ndata = preds.Size();
out_gpair->Resize(ndata);
@@ -81,8 +81,7 @@ class RegLossObj : public ObjFunction {
bool on_device = device >= 0;
// On CPU we run the transformation each thread processing a contigious block of data
// for better performance.
const size_t n_data_blocks =
std::max(static_cast<size_t>(1), (on_device ? ndata : nthreads));
const size_t n_data_blocks = std::max(static_cast<size_t>(1), (on_device ? ndata : nthreads));
const size_t block_size = ndata / n_data_blocks + !!(ndata % n_data_blocks);
common::Transform<>::Init(
[block_size, ndata] XGBOOST_DEVICE(
@@ -116,7 +115,7 @@ class RegLossObj : public ObjFunction {
}
},
common::Range{0, static_cast<int64_t>(n_data_blocks)}, device)
.Eval(&additional_input_, out_gpair, &preds, &info.labels_,
.Eval(&additional_input_, out_gpair, &preds, info.labels.Data(),
&info.weights_);
auto const flag = additional_input_.HostVector().begin()[0];
@@ -218,8 +217,8 @@ class PoissonRegression : public ObjFunction {
void GetGradient(const HostDeviceVector<bst_float>& preds,
const MetaInfo &info, int,
HostDeviceVector<GradientPair> *out_gpair) override {
CHECK_NE(info.labels_.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels_.Size()) << "labels are not correctly provided";
CHECK_NE(info.labels.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels.Size()) << "labels are not correctly provided";
size_t const ndata = preds.Size();
out_gpair->Resize(ndata);
auto device = tparam_->gpu_id;
@@ -249,7 +248,7 @@ class PoissonRegression : public ObjFunction {
expf(p + max_delta_step) * w};
},
common::Range{0, static_cast<int64_t>(ndata)}, device).Eval(
&label_correct_, out_gpair, &preds, &info.labels_, &info.weights_);
&label_correct_, out_gpair, &preds, info.labels.Data(), &info.weights_);
// copy "label correct" flags back to host
std::vector<int>& label_correct_h = label_correct_.HostVector();
for (auto const flag : label_correct_h) {
@@ -313,8 +312,8 @@ class CoxRegression : public ObjFunction {
void GetGradient(const HostDeviceVector<bst_float>& preds,
const MetaInfo &info, int,
HostDeviceVector<GradientPair> *out_gpair) override {
CHECK_NE(info.labels_.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels_.Size()) << "labels are not correctly provided";
CHECK_NE(info.labels.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels.Size()) << "labels are not correctly provided";
const auto& preds_h = preds.HostVector();
out_gpair->Resize(preds_h.size());
auto& gpair = out_gpair->HostVector();
@@ -334,7 +333,7 @@ class CoxRegression : public ObjFunction {
}
// start calculating grad and hess
const auto& labels = info.labels_.HostVector();
const auto& labels = info.labels.HostView();
double r_k = 0;
double s_k = 0;
double last_exp_p = 0.0;
@@ -345,7 +344,7 @@ class CoxRegression : public ObjFunction {
const double p = preds_h[ind];
const double exp_p = std::exp(p);
const double w = info.GetWeight(ind);
const double y = labels[ind];
const double y = labels(ind);
const double abs_y = std::abs(y);
// only update the denominator after we move forward in time (labels are sorted)
@@ -414,8 +413,8 @@ class GammaRegression : public ObjFunction {
void GetGradient(const HostDeviceVector<bst_float> &preds,
const MetaInfo &info, int,
HostDeviceVector<GradientPair> *out_gpair) override {
CHECK_NE(info.labels_.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels_.Size()) << "labels are not correctly provided";
CHECK_NE(info.labels.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels.Size()) << "labels are not correctly provided";
const size_t ndata = preds.Size();
auto device = tparam_->gpu_id;
out_gpair->Resize(ndata);
@@ -443,7 +442,7 @@ class GammaRegression : public ObjFunction {
_out_gpair[_idx] = GradientPair((1 - y / expf(p)) * w, y / expf(p) * w);
},
common::Range{0, static_cast<int64_t>(ndata)}, device).Eval(
&label_correct_, out_gpair, &preds, &info.labels_, &info.weights_);
&label_correct_, out_gpair, &preds, info.labels.Data(), &info.weights_);
// copy "label correct" flags back to host
std::vector<int>& label_correct_h = label_correct_.HostVector();
@@ -514,8 +513,8 @@ class TweedieRegression : public ObjFunction {
void GetGradient(const HostDeviceVector<bst_float>& preds,
const MetaInfo &info, int,
HostDeviceVector<GradientPair> *out_gpair) override {
CHECK_NE(info.labels_.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels_.Size()) << "labels are not correctly provided";
CHECK_NE(info.labels.Size(), 0U) << "label set cannot be empty";
CHECK_EQ(preds.Size(), info.labels.Size()) << "labels are not correctly provided";
const size_t ndata = preds.Size();
out_gpair->Resize(ndata);
@@ -550,7 +549,7 @@ class TweedieRegression : public ObjFunction {
_out_gpair[_idx] = GradientPair(grad * w, hess * w);
},
common::Range{0, static_cast<int64_t>(ndata), 1}, device)
.Eval(&label_correct_, out_gpair, &preds, &info.labels_, &info.weights_);
.Eval(&label_correct_, out_gpair, &preds, info.labels.Data(), &info.weights_);
// copy "label correct" flags back to host
std::vector<int>& label_correct_h = label_correct_.HostVector();