mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 12:21:27 +01:00
a9b0a921d5
Summary: As GoogleTest `TEST` macro is non-compliant with it as well as `DEFINE_DISPATCH` All changes but the ones to `.clang-tidy` are generated using following script: ``` for i in `find . -type f -iname "*.c*" -or -iname "*.h"|xargs grep cppcoreguidelines-avoid-non-const-global-variables|cut -f1 -d:|sort|uniq`; do sed -i "/\/\/ NOLINTNEXTLINE(cppcoreguidelines-avoid-non-const-global-variables)/d" $i; done ``` Pull Request resolved: https://github.com/pytorch/pytorch/pull/62008 Reviewed By: driazati, r-barnes Differential Revision: D29838584 Pulled By: malfet fbshipit-source-id: 1b2f8602c945bd4ce50a9bfdd204755556e31d13
104 lines
2.9 KiB
C++
104 lines
2.9 KiB
C++
#include "caffe2/operators/logit_op.h"
|
|
|
|
#include <string>
|
|
#include <vector>
|
|
|
|
#include "caffe2/operators/elementwise_ops.h"
|
|
#include "caffe2/utils/eigen_utils.h"
|
|
|
|
#include "caffe2/operators/bucketize_op.h"
|
|
|
|
#include "caffe2/core/operator.h"
|
|
#include "caffe2/core/tensor.h"
|
|
|
|
namespace caffe2 {
|
|
|
|
template <>
|
|
template <typename T>
|
|
bool LogitFunctor<CPUContext>::
|
|
operator()(const int size, const T* X, T* Y, CPUContext* /* context */) const {
|
|
ConstEigenVectorMap<T> X_vec(X, size);
|
|
EigenVectorMap<T> Y_vec(Y, size);
|
|
Y_vec = X_vec.array().min(static_cast<T>(1.0f - eps_));
|
|
Y_vec = Y_vec.array().max(eps_);
|
|
Y_vec = (Y_vec.array() / (T(1) - Y_vec.array())).log();
|
|
return true;
|
|
}
|
|
|
|
template <>
|
|
bool LogitGradientOp<float, CPUContext>::RunOnDevice() {
|
|
const auto& X = Input(0);
|
|
const auto& dY = Input(1);
|
|
|
|
auto* dX = Output(0, X.sizes(), at::dtype<float>());
|
|
int channels = X.dim32(X.dim() - 1);
|
|
ConstEigenArrayMap<float> Xmat(
|
|
X.template data<float>(), channels, X.numel() / channels);
|
|
ConstEigenArrayMap<float> dYmat(
|
|
dY.template data<float>(), channels, X.numel() / channels);
|
|
EigenArrayMap<float> dXmat(
|
|
dX->template mutable_data<float>(), channels, X.numel() / channels);
|
|
dXmat = (Xmat < eps_ || Xmat > 1.0 - eps_)
|
|
.select(0, dYmat * ((1 - Xmat) * Xmat).inverse());
|
|
return true;
|
|
}
|
|
|
|
REGISTER_CPU_OPERATOR(
|
|
Logit,
|
|
UnaryElementwiseWithArgsOp<
|
|
TensorTypes<float>,
|
|
CPUContext,
|
|
LogitFunctor<CPUContext>>);
|
|
|
|
REGISTER_CPU_OPERATOR(LogitGradient, LogitGradientOp<float, CPUContext>);
|
|
|
|
OPERATOR_SCHEMA(Logit)
|
|
.NumInputs(1)
|
|
.NumOutputs(1)
|
|
.AllowInplace({{0, 0}})
|
|
.IdenticalTypeAndShape()
|
|
.SetDoc(R"DOC(
|
|
Elementwise logit transform: logit(x) = log(x / (1 - x)), where x is the
|
|
input data clampped in (eps, 1-eps).
|
|
)DOC")
|
|
.Arg("eps (optional)", "small positive epsilon value, the default is 1e-6.")
|
|
.Input(0, "X", "input float tensor")
|
|
.Output(0, "Y", "output float tensor");
|
|
|
|
OPERATOR_SCHEMA(LogitGradient)
|
|
.NumInputs(2)
|
|
.NumOutputs(1)
|
|
.Input(0, "X", "input float tensor")
|
|
.Input(1, "dY", "input float tensor")
|
|
.Output(0, "dX", "output float tensor")
|
|
.Arg("eps", "small positive epsilon value, the default is 1e-6.");
|
|
|
|
namespace {
|
|
|
|
class GetLogitGradient : public GradientMakerBase {
|
|
using GradientMakerBase::GradientMakerBase;
|
|
vector<OperatorDef> GetGradientDefs() override {
|
|
return vector<OperatorDef>{CreateOperatorDef(
|
|
"LogitGradient",
|
|
"",
|
|
std::vector<std::string>{I(0), GO(0)},
|
|
std::vector<std::string>{GI(0)})};
|
|
}
|
|
};
|
|
|
|
} // namespace
|
|
|
|
REGISTER_GRADIENT(Logit, GetLogitGradient);
|
|
|
|
} // namespace caffe2
|
|
|
|
using LogitOp = caffe2::UnaryElementwiseWithArgsOp<
|
|
caffe2::TensorTypes<float>,
|
|
caffe2::CPUContext,
|
|
caffe2::LogitFunctor<caffe2::CPUContext>>;
|
|
|
|
C10_EXPORT_CAFFE2_OP_TO_C10_CPU(
|
|
Logit,
|
|
"_caffe2::Logit(Tensor X, float eps = 1e-6)->Tensor Y",
|
|
LogitOp);
|