nce_op.cc

/* 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/nce_op.h"

namespace paddle {
namespace operators {

using framework::Tensor;

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

  void InferShape(framework::InferShapeContext* ctx) const override {
    PADDLE_ENFORCE(ctx->HasInput("Input"));
    PADDLE_ENFORCE(ctx->HasInput("Label"));
    PADDLE_ENFORCE(ctx->HasInput("Weight"));
    PADDLE_ENFORCE(ctx->HasOutput("Cost"));
    PADDLE_ENFORCE(ctx->HasOutput("SampleLogits"));
    PADDLE_ENFORCE(ctx->HasOutput("SampleLabels"));

    auto x_dims = ctx->GetInputDim("Input");
    auto label_dims = ctx->GetInputDim("Label");
    PADDLE_ENFORCE_EQ(x_dims[0], label_dims[0]);
    int num_true_classes = label_dims.size() == 2 ? label_dims[1] : 1;
    if (ctx->HasInput("Bias")) {
      PADDLE_ENFORCE_EQ(ctx->GetInputDim("Weight")[0],
                        ctx->GetInputDim("Bias")[0]);
    }
    auto num_sampled_classes = ctx->Attrs().Get<int>("num_sampled_classes");
    auto num_classes = ctx->Attrs().Get<int>("num_classes");
    std::vector<int> sampled_labels =
        ctx->Attrs().Get<std::vector<int>>("sampled_labels");
    PADDLE_ENFORCE_EQ(num_classes, ctx->GetInputDim("Weight")[0]);
    PADDLE_ENFORCE_LT(num_sampled_classes, num_classes);
    if (sampled_labels.size() > 0) {
      PADDLE_ENFORCE_EQ(sampled_labels.size(),
                        static_cast<size_t>(num_sampled_classes));
    }
    // set dims of output(Out)
    std::vector<int64_t> out_dims;
    out_dims.push_back(x_dims[0]);
    ctx->SetOutputDim("Cost", framework::make_ddim(out_dims));

    // set dims of output(SampleOut)
    std::vector<int64_t> sample_out_dims;
    sample_out_dims.push_back(x_dims[0]);
    sample_out_dims.push_back(num_sampled_classes + num_true_classes);
    ctx->SetOutputDim("SampleLogits", framework::make_ddim(sample_out_dims));
    ctx->SetOutputDim("SampleLabels", framework::make_ddim(sample_out_dims));
  }

 protected:
  framework::OpKernelType GetKernelType(
      const framework::ExecutionContext& ctx) const override {
    return framework::OpKernelType(
        framework::ToDataType(ctx.Input<Tensor>("Input")->type()),
        ctx.device_context());
  }
};

class NCEOpMaker : public framework::OpProtoAndCheckerMaker {
 public:
  NCEOpMaker(framework::OpProto* proto, framework::OpAttrChecker* op_checker)
      : OpProtoAndCheckerMaker(proto, op_checker) {
    AddInput("Input", "(Tensor) A tensor of shape [batch_size, dim].");
    AddInput("Label",
             "(Tensor) A tensor of shape [batch_size, num_true_class]. "
             "'num_true_class' is the number of target class in each sample.");
    AddInput("Weight",
             "(Tensor) A tensor of shape [num_class, dim]. 'num_class' is the "
             "total number of class.");
    AddInput("Bias",
             "(Tensor) A tensor of shape [num_class]. 'num_class' is the total "
             "number of class. It is a dispensable input.")
        .AsDispensable();
    AddInput("SampleWeight",
             "(Tensor) A tensor of shape [batch_size] storing a weight for "
             "each sample. And it is a dispensable input. The default value of "
             "sample is 1.")
        .AsDispensable();
    AddOutput("Cost",
              "(Tensor) A tensor of shape [batch_size]. Cost of samples.");
    AddOutput("SampleLogits", "An intermediate tensor.").AsIntermediate();
    AddOutput("SampleLabels", "An intermediate tensor.").AsIntermediate();
    AddAttr<int>("num_classes", "Total number of classes.");
    AddAttr<int>("num_sampled_classes", "The number of negative classes.")
        .SetDefault(10);
    AddAttr<std::vector<int>>("sampled_labels", "");
    AddComment(R"DOC(
Computes and returns the noise-contrastive estimation training loss.
See [Noise-contrastive estimation: A new estimation principle for unnormalized statistical models](http://www.jmlr.org/proceedings/papers/v9/gutmann10a/gutmann10a.pdf).
By default this uses a uniform distribution for sampling.
The number of target classes per example should be same. If you have a variable number of target classes, you can pad them out to a constant number by either repeating them or by padding with an otherwise unused class.
)DOC");
  }
};

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

  void InferShape(framework::InferShapeContext* ctx) const override {
    PADDLE_ENFORCE(ctx->HasInput("Input"));
    PADDLE_ENFORCE(ctx->HasInput("Weight"));
    PADDLE_ENFORCE(ctx->HasInput("Cost"));
    PADDLE_ENFORCE(ctx->HasInput("SampleLogits"));
    PADDLE_ENFORCE(ctx->HasInput("SampleLabels"));
    PADDLE_ENFORCE(ctx->HasInput(framework::GradVarName("Cost")),
                   "The input(Out@GRAD) should not be null");

    auto x_dims = ctx->GetInputDim("Input");
    auto x_grad_name = framework::GradVarName("Input");
    if (ctx->HasOutput(x_grad_name)) {
      ctx->SetOutputDim(x_grad_name, x_dims);
    }

    auto w_dims = ctx->GetInputDim("Weight");
    auto w_grad_name = framework::GradVarName("Weight");
    if (ctx->HasOutput(w_grad_name)) {
      ctx->SetOutputDim(w_grad_name, w_dims);
    }

    auto bias_grad_name = framework::GradVarName("Bias");
    if (ctx->HasOutput(bias_grad_name)) {
      auto bias_dims = ctx->GetInputDim("Bias");
      ctx->SetOutputDim(bias_grad_name, bias_dims);
    }
  }

 protected:
  framework::OpKernelType GetKernelType(
      const framework::ExecutionContext& ctx) const override {
    return framework::OpKernelType(
        framework::ToDataType(ctx.Input<Tensor>("Input")->type()),
        ctx.device_context());
  }
};

}  // namespace operators
}  // namespace paddle

namespace ops = paddle::operators;
REGISTER_OP(nce, ops::NCEOp, ops::NCEOpMaker, nce_grad, ops::NCEOpGrad);
REGISTER_OP_CPU_KERNEL(nce, ops::NCEKernel<paddle::platform::CPUPlace, float>);
REGISTER_OP_CPU_KERNEL(nce_grad,
                       ops::NCEGradKernel<paddle::platform::CPUPlace, float>);