dangqingqing
8 years ago
parent
d11430e009
commit
f299206396
@ -0,0 +1,90 @@
|
||||
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
|
||||
|
||||
Licensed under the Apache License, Version 2.0 (the "License");
|
||||
you may not use this file except in compliance with the License.
|
||||
You may obtain a copy of the License at
|
||||
|
||||
http://www.apache.org/licenses/LICENSE-2.0
|
||||
|
||||
Unless required by applicable law or agreed to in writing, software
|
||||
distributed under the License is distributed on an "AS IS" BASIS,
|
||||
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
See the License for the specific language governing permissions and
|
||||
limitations under the License. */
|
||||
|
||||
#include "paddle/operators/sequence_avg_pool_op.h"
|
||||
|
||||
namespace paddle {
|
||||
namespace operators {
|
||||
|
||||
class SequenceAvgPoolOp : public framework::OperatorWithKernel {
|
||||
public:
|
||||
using framework::OperatorWithKernel::OperatorWithKernel;
|
||||
|
||||
protected:
|
||||
void InferShape(const framework::InferShapeContext& ctx) const override {
|
||||
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("X"),
|
||||
"Input of SequenceAvgPoolOp"
|
||||
"must be initialized.");
|
||||
auto* x = ctx.Input<framework::LoDTensor>("X");
|
||||
auto dims = x->dims();
|
||||
auto lod = x->lod();
|
||||
PADDLE_ENFORCE_EQ(lod.size(), 1UL, "Only support one level sequence now.");
|
||||
PADDLE_ENFORCE_GE(
|
||||
dims[0],
|
||||
/*batch size = */ static_cast<int64_t>(lod[0].size() - 1),
|
||||
"The first dimension of Input(X) must be large than batch size.");
|
||||
dims[0] = lod[0].size() - 1;
|
||||
ctx.Output<framework::LoDTensor>("Out")->Resize({dims});
|
||||
}
|
||||
};
|
||||
|
||||
class SequenceAvgPoolOpMaker : public framework::OpProtoAndCheckerMaker {
|
||||
public:
|
||||
SequenceAvgPoolOpMaker(framework::OpProto* proto,
|
||||
framework::OpAttrChecker* op_checker)
|
||||
: OpProtoAndCheckerMaker(proto, op_checker) {
|
||||
AddInput("X", "Input of SequenceAvgPoolOp.");
|
||||
AddOutput("Out", "The output of SequenceAvgPoolOp.");
|
||||
AddComment(R"DOC(
|
||||
SequenceAvgPoolOp averages features of all time-steps of each instance.
|
||||
More detailed comments will be added later.
|
||||
)DOC");
|
||||
}
|
||||
};
|
||||
|
||||
class SequenceAvgPoolGradOp : public framework::OperatorWithKernel {
|
||||
public:
|
||||
using framework::OperatorWithKernel::OperatorWithKernel;
|
||||
|
||||
protected:
|
||||
void InferShape(const framework::InferShapeContext& ctx) const override {
|
||||
PADDLE_ENFORCE_NOT_NULL(ctx.InputVar(framework::GradVarName("Out")),
|
||||
"Gradient of Out should not be null");
|
||||
auto og_dims =
|
||||
ctx.Input<framework::LoDTensor>(framework::GradVarName("Out"))->dims();
|
||||
auto x_dims = ctx.Input<framework::LoDTensor>("X")->dims();
|
||||
PADDLE_ENFORCE_EQ(og_dims.size(), x_dims.size(),
|
||||
"The rank of output grad must equal to Input(X).");
|
||||
for (size_t i = 1; i < og_dims.size(); ++i) {
|
||||
PADDLE_ENFORCE_EQ(og_dims[i], x_dims[i], "The dimension mismatch.");
|
||||
}
|
||||
auto* x_grad =
|
||||
ctx.Output<framework::LoDTensor>(framework::GradVarName("X"));
|
||||
x_grad->Resize(x_dims);
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace operators
|
||||
} // namespace paddle
|
||||
|
||||
namespace ops = paddle::operators;
|
||||
REGISTER_OP(sequence_avg_pool, ops::SequenceAvgPoolOp,
|
||||
ops::SequenceAvgPoolOpMaker, sequence_avg_pool_grad,
|
||||
ops::SequenceAvgPoolGradOp);
|
||||
REGISTER_OP_CPU_KERNEL(
|
||||
sequence_avg_pool,
|
||||
ops::SequenceAvgPoolKernel<paddle::platform::CPUPlace, float>);
|
||||
REGISTER_OP_CPU_KERNEL(
|
||||
sequence_avg_pool_grad,
|
||||
ops::SequenceAvgPoolGradKernel<paddle::platform::CPUPlace, float>);
|
||||
@ -0,0 +1,25 @@
|
||||
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
|
||||
|
||||
Licensed under the Apache License, Version 2.0 (the "License");
|
||||
you may not use this file except in compliance with the License.
|
||||
You may obtain a copy of the License at
|
||||
|
||||
http://www.apache.org/licenses/LICENSE-2.0
|
||||
|
||||
Unless required by applicable law or agreed to in writing, software
|
||||
distributed under the License is distributed on an "AS IS" BASIS,
|
||||
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
See the License for the specific language governing permissions and
|
||||
limitations under the License. */
|
||||
|
||||
#define EIGEN_USE_GPU
|
||||
|
||||
#include "paddle/operators/sequence_avg_pool_op.h"
|
||||
|
||||
namespace ops = paddle::operators;
|
||||
REGISTER_OP_GPU_KERNEL(
|
||||
sequence_avg_pool,
|
||||
ops::SequenceAvgPoolKernel<paddle::platform::GPUPlace, float>);
|
||||
REGISTER_OP_GPU_KERNEL(
|
||||
sequence_avg_pool_grad,
|
||||
ops::SequenceAvgPoolGradKernel<paddle::platform::GPUPlace, float>);
|
||||
@ -0,0 +1,81 @@
|
||||
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
|
||||
|
||||
Licensed under the Apache License, Version 2.0 (the "License");
|
||||
you may not use this file except in compliance with the License.
|
||||
You may obtain a copy of the License at
|
||||
|
||||
http://www.apache.org/licenses/LICENSE-2.0
|
||||
|
||||
Unless required by applicable law or agreed to in writing, software
|
||||
distributed under the License is distributed on an "AS IS" BASIS,
|
||||
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
See the License for the specific language governing permissions and
|
||||
limitations under the License. */
|
||||
|
||||
#pragma once
|
||||
#include "paddle/framework/eigen.h"
|
||||
#include "paddle/framework/op_registry.h"
|
||||
|
||||
namespace paddle {
|
||||
namespace operators {
|
||||
|
||||
using Tensor = framework::Tensor;
|
||||
using LoDTensor = framework::LoDTensor;
|
||||
template <typename T, int MajorType = Eigen::RowMajor,
|
||||
typename IndexType = Eigen::DenseIndex>
|
||||
using EigenMatrix = framework::EigenMatrix<T, MajorType, IndexType>;
|
||||
|
||||
template <typename Place, typename T>
|
||||
class SequenceAvgPoolKernel : public framework::OpKernel {
|
||||
public:
|
||||
void Compute(const framework::ExecutionContext& context) const override {
|
||||
auto* in = context.Input<LoDTensor>("X");
|
||||
auto* out = context.Output<LoDTensor>("Out");
|
||||
|
||||
auto dims = in->dims();
|
||||
auto lod = in->lod();
|
||||
int64_t w = in->numel() / dims[0];
|
||||
|
||||
out->mutable_data<T>(context.GetPlace());
|
||||
auto place = context.GetEigenDevice<Place>();
|
||||
for (int i = 0; i < lod[0].size() - 1; ++i) {
|
||||
Tensor in_t = in->Slice<T>(static_cast<int>(lod[0][i]),
|
||||
static_cast<int>(lod[0][i + 1]));
|
||||
Tensor out_t = out->Slice<T>(i, i + 1);
|
||||
int64_t h = static_cast<int64_t>(lod[0][i + 1] - lod[0][i]);
|
||||
auto in_e = EigenMatrix<T>::From(in_t, {h, w});
|
||||
auto out_e = EigenMatrix<T>::From(out_t, {h, w});
|
||||
out_e.device(place) = in_e.mean(Eigen::array<int, 1>({{0}}));
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
template <typename Place, typename T>
|
||||
class SequenceAvgPoolGradKernel : public framework::OpKernel {
|
||||
public:
|
||||
void Compute(const framework::ExecutionContext& context) const override {
|
||||
auto* in = context.Output<LoDTensor>("X");
|
||||
auto* in_g = context.Output<LoDTensor>(framework::GradVarName("X"));
|
||||
auto* out_g = context.Input<LoDTensor>(framework::GradVarName("Out"));
|
||||
|
||||
auto dims = in->dims();
|
||||
auto lod = in->lod();
|
||||
int64_t w = in->numel() / dims[0];
|
||||
|
||||
in_g->mutable_data<T>(context.GetPlace());
|
||||
auto place = context.GetEigenDevice<Place>();
|
||||
for (int i = 0; i < lod[0].size() - 1; ++i) {
|
||||
auto in_g_t = in_g->Slice<T>(static_cast<int>(lod[0][i]),
|
||||
static_cast<int>(lod[0][i + 1]));
|
||||
auto out_g_t = out_g->Slice<T>(i, i + 1);
|
||||
int64_t h = static_cast<int64_t>(lod[0][i + 1] - lod[0][i]);
|
||||
auto in_g_e = EigenMatrix<T>::From(in_g_t, {h, w});
|
||||
auto out_g_e = EigenMatrix<T>::From(out_g_t, {1, w});
|
||||
Eigen::DSizes<int, 2> bcast(h, w);
|
||||
in_g_e.device(place) = (out_g_e / static_cast<T>(h)).broadcast(bcast);
|
||||
}
|
||||
}
|
||||
};
|
||||
|
||||
} // namespace operators
|
||||
} // namespace paddle
|
||||
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in new issue