dropout_op.cc 4.4 KB
Newer Older
X
Xinghai Sun 已提交
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
/* 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/dropout_op.h"

namespace paddle {
namespace operators {

using framework::Tensor;
21
using framework::LoDTensor;
X
Xinghai Sun 已提交
22 23 24 25 26 27 28

class DropoutOp : public framework::OperatorWithKernel {
 public:
  using framework::OperatorWithKernel::OperatorWithKernel;

 protected:
  void InferShape(const framework::InferShapeContext &ctx) const override {
29
    // validity check
X
Xinghai Sun 已提交
30
    PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("X"), "Input(X) must not be null.");
31 32
    PADDLE_ENFORCE_GE(ctx.Attr<float>("dropout_prob"), 0);
    PADDLE_ENFORCE_LE(ctx.Attr<float>("dropout_prob"), 1);
33 34 35 36
    // TODO(xinghai-sun): remove this check after swtiching to bool
    PADDLE_ENFORCE(ctx.Attr<int>("is_training") == 0 ||
                   ctx.Attr<int>("is_training") == 1);

37
    // resize
X
Xinghai Sun 已提交
38
    auto dims = ctx.Input<Tensor>("X")->dims();
39 40
    ctx.Output<LoDTensor>("Out")->Resize(dims);
    ctx.Output<LoDTensor>("Mask")->Resize(dims);
X
Xinghai Sun 已提交
41 42 43
  }
};

44
template <typename AttrType>
X
Xinghai Sun 已提交
45 46 47 48 49
class DropoutOpMaker : public framework::OpProtoAndCheckerMaker {
 public:
  DropoutOpMaker(framework::OpProto *proto,
                 framework::OpAttrChecker *op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
50
    AddAttr<AttrType>("dropout_prob", "Probability of setting units to zero.")
51
        .SetDefault(.5f);
52 53
    // TODO(xinghai-sun): use bool for is_training after bool is supported.
    AddAttr<int>("is_training", "Whether in training phase.").SetDefault(1);
54
    AddAttr<int>("seed", "Dropout random seed.").SetDefault(0);
X
Xinghai Sun 已提交
55 56
    AddInput("X", "The input of dropout op.");
    AddOutput("Out", "The output of dropout op.");
57
    AddOutput("Mask", "The random sampled dropout mask.").AsIntermediate();
X
Xinghai Sun 已提交
58

59 60 61 62 63 64
    AddComment(R"DOC(
Dropout Operator.

"Dropout" refers to randomly dropping out units in a nerual network. It is a
regularization technique for reducing overfitting by preventing neuron
co-adaption during training. The dropout operator randomly set (according to
65
the given dropout probability) the outputs of some units to zero, while others
66 67
being set to their inputs.
)DOC");
X
Xinghai Sun 已提交
68 69 70
  }
};

71
template <typename AttrType>
X
Xinghai Sun 已提交
72 73 74 75 76 77
class DropoutOpGrad : public framework::OperatorWithKernel {
 public:
  using framework::OperatorWithKernel::OperatorWithKernel;

 protected:
  void InferShape(const framework::InferShapeContext &ctx) const override {
78
    // validity check
X
Xinghai Sun 已提交
79 80 81 82
    PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("X"), "Input(X) must not be null.");
    PADDLE_ENFORCE_NOT_NULL(ctx.InputVar("Mask"), "Mask must not be null.");
    PADDLE_ENFORCE_NOT_NULL(ctx.InputVar(framework::GradVarName("Out")),
                            "Input(Out@GRAD) must not be null.");
83 84 85 86 87
    PADDLE_ENFORCE_GE(ctx.Attr<AttrType>("dropout_prob"), 0);
    PADDLE_ENFORCE_LE(ctx.Attr<AttrType>("dropout_prob"), 1);
    // TODO(xinghai-sun): remove this check after swtiching to bool
    PADDLE_ENFORCE(ctx.Attr<int>("is_training") == 0 ||
                   ctx.Attr<int>("is_training") == 1);
X
Xinghai Sun 已提交
88 89 90 91
    auto x_dims = ctx.Input<Tensor>("X")->dims();
    auto mask_dims = ctx.Input<Tensor>("Mask")->dims();
    auto out_dims = ctx.Input<Tensor>(framework::GradVarName("Out"))->dims();
    PADDLE_ENFORCE_EQ(x_dims, out_dims,
X
Xinghai Sun 已提交
92
                      "Dimensions of Input(X) and Out@Grad must be the same.");
X
Xinghai Sun 已提交
93 94
    PADDLE_ENFORCE_EQ(x_dims, mask_dims,
                      "Dimensions of Input(X) and Mask must be the same.");
95
    // resize
96
    auto *x_grad = ctx.Output<LoDTensor>(framework::GradVarName("X"));
X
Xinghai Sun 已提交
97 98 99 100 101 102 103 104
    x_grad->Resize(x_dims);
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
105 106
REGISTER_OP(dropout, ops::DropoutOp, ops::DropoutOpMaker<float>, dropout_grad,
            ops::DropoutOpGrad<float>);
107
REGISTER_OP_CPU_KERNEL(
108
    dropout, ops::CPUDropoutKernel<paddle::platform::CPUPlace, float, float>);
X
Xinghai Sun 已提交
109 110
REGISTER_OP_CPU_KERNEL(
    dropout_grad, ops::DropoutGradKernel<paddle::platform::CPUPlace, float>);