ContextProjectionOp.cpp

/* 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 "ContextProjectionOp.h"
#include "paddle/math/Matrix.h"
#include "paddle/math/Vector.h"

namespace paddle {
/**
 * Context Projection Forward with CPU Matrix Device.
 *
 */
template <>
void ContextProjectionForward<DEVICE_TYPE_CPU>(CpuMatrix& out_mat,
                                               const CpuMatrix& input_mat,
                                               const CpuMatrix& weight_mat,
                                               const CpuIVector& seq_vec,
                                               size_t context_length,
                                               int context_start,
                                               size_t begin_pad) {
  const int* starts = seq_vec.getData();
  const size_t num_sequences = seq_vec.getSize() - 1;
  for (size_t i = 0; i < num_sequences; ++i) {
    for (size_t j = 0; j < context_length; ++j) {
      int begin = starts[i] + context_start + j;
      int end = starts[i + 1] + context_start + j;
      int dst_begin = starts[i];
      int dst_end = starts[i + 1];
      if (begin < starts[i]) {
        int64_t pad_size =
            std::min(starts[i] - begin, starts[i + 1] - starts[i]);
        MatrixPtr mat = out_mat.subMatrix(starts[i], pad_size);
        if (weight_mat) {
          MatrixPtr sub =
              const_cast<CpuMatrix&>(weight_mat).subMatrix(j, pad_size);
          mat->addAtOffset(*sub, j * input_mat.getWidth());
        }
        dst_begin = starts[i] + pad_size;
        begin = starts[i];
      }
      if (end > starts[i + 1]) {
        int64_t pad_size =
            std::min(end - starts[i + 1], starts[i + 1] - starts[i]);
        MatrixPtr mat = out_mat.subMatrix(starts[i + 1] - pad_size, pad_size);
        if (weight_mat) {
          MatrixPtr sub =
              const_cast<CpuMatrix&>(weight_mat)
                  .subMatrix(begin_pad + context_start + j - pad_size,
                             pad_size);
          mat->addAtOffset(*sub, j * input_mat.getWidth());
        }
        dst_end = starts[i + 1] - pad_size;
        end = starts[i + 1];
      }
      if (end <= begin) continue;
      MatrixPtr src =
          const_cast<CpuMatrix&>(input_mat).subMatrix(begin, end - begin);
      MatrixPtr dst = out_mat.subMatrix(dst_begin, dst_end - dst_begin);
      dst->addAtOffset(*src, j * input_mat.getWidth());
    }
  }
}

/**
 * Paddle Function for Context Projection Forward.
 * Calculate the output sequence after context projection.
 *
 * What is Context Projection?
 * For example, assumed input (x) has 4 words and the dimension of each word
 * representation is 2. If we use zero to pad instead of learned weight to pad,
 * and the context_lenth is 3, the output (y) is:
 *
 * @code
 *  x = [a1, a2;
 *       b1, b2;
 *       c1, c2;
 *       d1, d2]
 *  y = [0,  0,  a1, a2, b1, b2;
 *       a1, a2, b1, b2, c1, c2;
 *       b1, b2, c1, c2, d1, d2;
 *       c1, c2, d1, d2, 0,  0]
 * @endcode
 *
 * \param outputs[0].matrix   output value, n * (d * l)
 * \param outputs[0].vector   input sequence, n * 1
 * \param inputs[0].matrix    input value, n * d
 * \param inputs[0].vector    input sequence, n * 1
 * \param inputs[1].matrix    input weight, pad * d
 * \param inputs[1].vector    input sequence, n * 1
 */
template <DeviceType Device>
class ContextProjectionForwardFunc : public FunctionBase {
public:
  void init(const FuncConfig& config) override {
    context_length_ = config.get<size_t>("context_length");
    context_start_ = config.get<int>("context_start");
    begin_pad_ = config.get<size_t>("begin_pad");
  }

  void calc(const BufferArgs& inputs, const BufferArgs& outputs) override {
    CHECK(1 == inputs.size() || 2 == inputs.size());
    CHECK_EQ((size_t)1, outputs.size());

    const auto val_seqs = dynamic_cast<const SequenceArg&>(inputs[0]);
    const auto w_seqs = inputs.size() <= 1
                            ? nullptr
                            : dynamic_cast<const SequenceArg*>(&inputs[1]);
    auto out_seqs = dynamic_cast<const SequenceArg&>(outputs[0]);

    CHECK(out_seqs.data() && val_seqs.data() &&
          val_seqs.getSequenceIds().data());
    CHECK_EQ(out_seqs.shape().ndims(), (size_t)2);
    CHECK_EQ(val_seqs.shape().ndims(), (size_t)2);
    CHECK_EQ(val_seqs.getSequenceIds().shape().ndims(), (size_t)1);
    if (w_seqs) {
      CHECK_EQ(w_seqs->shape().ndims(), (size_t)2);
      CHECK_EQ(w_seqs->getSequenceIds().shape().ndims(), (size_t)1);
    }
    /// dim of output = dim of input * context_length
    CHECK_EQ(out_seqs.shape()[1], val_seqs.shape()[1] * context_length_);
    /// input and output has the same batch_size
    CHECK_EQ(val_seqs.shape()[0], out_seqs.shape()[0]);
    /// dim of input == dim of weight
    if (w_seqs) {
      CHECK_EQ(val_seqs.shape()[1], w_seqs->shape()[1]);
    }

    CHECK_EQ(out_seqs.getArgType(), ADD_TO);
    auto out_mat = out_seqs.matrix<Device>();
    const auto in_mat = val_seqs.matrix<Device>();
    const auto w_mat =
        w_seqs ? w_seqs->matrix<Device>()
               : typename Tensor<real, Device>::Matrix(nullptr, 0, 0);
    const auto seq_vec = val_seqs.getSequenceIds().vector<int, Device>();
    ContextProjectionForward<Device>(out_mat,
                                     in_mat,
                                     w_mat,
                                     seq_vec,
                                     context_length_,
                                     context_start_,
                                     begin_pad_);
  }

private:
  size_t context_length_;
  int context_start_;
  size_t begin_pad_;
};

/**
 * Context Projection Backward with CPU Matrix Device.
 *
 */
template <>
void ContextProjectionBackward<DEVICE_TYPE_CPU>(const CpuMatrix& out_grad_mat,
                                                CpuMatrix& in_grad_mat,
                                                CpuMatrix& w_grad_mat,
                                                const CpuIVector& seq_vec,
                                                size_t context_length,
                                                int context_start,
                                                size_t begin_pad,
                                                bool is_padding,
                                                size_t total_pad) {
  size_t input_dim = in_grad_mat ? in_grad_mat.getWidth()
                                 : w_grad_mat ? w_grad_mat.getWidth() : 0;
  const int* starts = seq_vec.getData();
  size_t num_sequences = seq_vec.getSize() - 1;
  for (size_t i = 0; i < num_sequences; ++i) {
    for (size_t j = 0; j < context_length; ++j) {
      int begin = starts[i] + context_start + j;
      int end = starts[i + 1] + context_start + j;
      int dst_begin = starts[i];
      int dst_end = starts[i + 1];
      if (begin < starts[i]) {
        int64_t pad_size =
            std::min(starts[i] - begin, starts[i + 1] - starts[i]);
        if (is_padding && w_grad_mat) {
          MatrixPtr mat = const_cast<CpuMatrix&>(out_grad_mat)
                              .subMatrix(starts[i], pad_size);
          MatrixPtr sub = w_grad_mat.subMatrix(j, pad_size);
          sub->addAtOffset(*mat, j * input_dim);
        }
        dst_begin = starts[i] + pad_size;
        begin = starts[i];
      }
      if (end > starts[i + 1]) {
        int64_t pad_size =
            std::min(end - starts[i + 1], starts[i + 1] - starts[i]);
        if (is_padding && w_grad_mat) {
          MatrixPtr mat = const_cast<CpuMatrix&>(out_grad_mat)
                              .subMatrix(starts[i + 1] - pad_size, pad_size);
          MatrixPtr sub = w_grad_mat.subMatrix(
              begin_pad + context_start + j - pad_size, pad_size);
          sub->addAtOffset(*mat, j * input_dim);
        }
        dst_end = starts[i + 1] - pad_size;
        end = starts[i + 1];
      }
      if (end <= begin) continue;
      if (!in_grad_mat) continue;
      MatrixPtr src = in_grad_mat.subMatrix(begin, end - begin);
      MatrixPtr dst = const_cast<CpuMatrix&>(out_grad_mat)
                          .subMatrix(dst_begin, dst_end - dst_begin);
      src->addAtOffset(*dst, j * input_dim);
    }
  }
}

/**
 * Context Projection Backward Function.
 * Update the weight gradient and input layer gradient with backprop
 *
 * \param inputs[0].seq          input sequence.
 * \param inputs[0].matrix       output layer grad.
 * \param outputs[0]             input layer grad.
 * \param outputs[1]             weight grad.
 */
template <DeviceType Device>
class ContextProjectionBackwardFunc : public FunctionBase {
public:
  void init(const FuncConfig& config) override {
    context_length_ = config.get<size_t>("context_length");
    context_start_ = config.get<int>("context_start");
    begin_pad_ = config.get<size_t>("begin_pad");
    is_padding_ = config.get<bool>("is_padding");
    total_pad_ = config.get<size_t>("total_pad");
  }

  void calc(const BufferArgs& inputs, const BufferArgs& outputs) override {
    CHECK_EQ((size_t)1, inputs.size());
    CHECK_EQ((size_t)2, outputs.size());

    const auto seq_arg = dynamic_cast<const SequenceArg&>(inputs[0]);
    CHECK(seq_arg.data() && inputs[0].data());
    CHECK_EQ(seq_arg.shape().ndims(), (size_t)2);
    CHECK_EQ(seq_arg.getSequenceIds().shape().ndims(), (size_t)1);
    CHECK_EQ(outputs[0].shape().ndims(), (size_t)2);
    CHECK_EQ(outputs[1].shape().ndims(), (size_t)2);

    /// dim of input grad == dim of weight
    CHECK_EQ(outputs[0].shape()[1], outputs[1].shape()[1]);
    /// input and output grad has the same batch_size
    CHECK_EQ(outputs[0].shape()[0], seq_arg.shape()[0]);
    /// dim of output val = dim of input grad * context_length
    CHECK_EQ(seq_arg.shape()[1], outputs[0].shape()[1] * context_length_);

    CHECK_EQ(outputs[0].getArgType(), ADD_TO);
    CHECK_EQ(outputs[1].getArgType(), ADD_TO);

    const auto seq_vec = seq_arg.getSequenceIds().vector<int, Device>();
    const auto out_grad_mat = seq_arg.matrix<Device>();
    auto in_grad_mat =
        !outputs[0].data()
            ? typename Tensor<real, Device>::Matrix(nullptr, 0, 0)
            : outputs[0].matrix<Device>();
    auto w_grad_mat = !outputs[1].data()
                          ? typename Tensor<real, Device>::Matrix(nullptr, 0, 0)
                          : outputs[1].matrix<Device>();
    ContextProjectionBackward<Device>(out_grad_mat,
                                      in_grad_mat,
                                      w_grad_mat,
                                      seq_vec,
                                      context_length_,
                                      context_start_,
                                      begin_pad_,
                                      is_padding_,
                                      total_pad_);
  }

private:
  size_t context_length_;
  int context_start_;
  size_t begin_pad_;
  bool is_padding_;
  size_t total_pad_;
};

/**
 * \param inputs[0].matrix      input grad, n*d
 * \param inputs[0].vector      input sequence, n*1
 * \param outputs[0]            output grad, n*(d*l)
 */
template <DeviceType Device>
class ContextProjectionBackwardDataFunc : public FunctionBase {
public:
  void init(const FuncConfig& config) override {
    context_length_ = config.get<size_t>("context_length");
    context_start_ = config.get<int>("context_start");
  }

  void calc(const BufferArgs& inputs, const BufferArgs& outputs) override {
    CHECK_EQ(1, static_cast<int>(inputs.size()));
    CHECK_EQ(1, static_cast<int>(outputs.size()));
    const auto in_seqs = dynamic_cast<const SequenceArg&>(inputs[0]);
    CHECK(in_seqs.data() && outputs[0].data() &&
          in_seqs.getSequenceIds().data());
    CHECK_EQ(static_cast<int>(outputs[0].shape().ndims()), 2);
    CHECK_EQ(static_cast<int>(in_seqs.shape().ndims()), 2);
    CHECK_EQ(static_cast<int>(in_seqs.getSequenceIds().shape().ndims()), 1);
    CHECK_EQ(outputs[0].shape().ndims(),
             in_seqs.shape().ndims() * context_length_);
    /// input and output has the same batch_size
    CHECK_EQ(in_seqs.shape()[0], outputs[0].shape()[0]);
    const auto out_grad_mat = outputs[0].matrix<Device>();
    auto in_grad_mat = in_seqs.matrix<Device>();
    const auto seq_vec = in_seqs.getSequenceIds().vector<int, Device>();

    ContextProjectionBackwardData<Device>(
        out_grad_mat, in_grad_mat, seq_vec, context_length_, context_start_);
  }

private:
  size_t context_length_;
  int context_start_;
};

/**
 * \param inputs[0].matrix    weight grad, pad * d
 * \param inputs[0].vecotr    input sequence, n * 1
 * \param outputs[0]          output grad, n * (d * l)
 */
template <DeviceType Device>
class ContextProjectionBackwardWeightFunc : public FunctionBase {
public:
  void init(const FuncConfig& config) override {
    context_length_ = config.get<size_t>("context_length");
    context_start_ = config.get<int>("context_start");
    begin_pad_ = config.get<size_t>("begin_pad");
    total_pad_ = config.get<size_t>("total_pad");
  }

  void calc(const BufferArgs& inputs, const BufferArgs& outputs) override {
    CHECK_EQ(1, static_cast<int>(inputs.size()));
    CHECK_EQ(1, static_cast<int>(outputs.size()));

    const auto in_seqs = dynamic_cast<const SequenceArg&>(inputs[0]);
    CHECK(in_seqs.data() && in_seqs.getSequenceIds().data() &&
          outputs[0].data());
    CHECK_EQ(static_cast<int>(outputs[0].shape().ndims()), 2);
    CHECK_EQ(static_cast<int>(in_seqs.shape().ndims()), 2);
    CHECK_EQ(static_cast<int>(in_seqs.getSequenceIds().shape().ndims()), 1);
    CHECK_EQ(in_seqs.shape()[0], outputs[0].shape()[0]);
    CHECK_EQ(outputs[0].shape()[1], in_seqs.shape()[1] * context_length_);
    const auto out_grad_mat = outputs[0].matrix<Device>();
    auto w_grad_mat = inputs[0].matrix<Device>();
    const auto seq_vec = in_seqs.getSequenceIds().vector<int, Device>();
    ContextProjectionBackwardWeight<Device>(out_grad_mat,
                                            w_grad_mat,
                                            seq_vec,
                                            context_length_,
                                            context_start_,
                                            total_pad_,
                                            begin_pad_);
  }

private:
  size_t context_length_;
  int context_start_;
  size_t begin_pad_;
  size_t total_pad_;
};

REGISTER_TYPED_FUNC(ContextProjectionForward,
                    CPU,
                    ContextProjectionForwardFunc);
REGISTER_TYPED_FUNC(ContextProjectionBackward,
                    CPU,
                    ContextProjectionBackwardFunc);
#ifndef PADDLE_ONLY_CPU
REGISTER_TYPED_FUNC(ContextProjectionForward,
                    GPU,
                    ContextProjectionForwardFunc);
REGISTER_TYPED_FUNC(ContextProjectionBackward,
                    GPU,
                    ContextProjectionBackwardFunc);
REGISTER_TYPED_FUNC(ContextProjectionBackwardData,
                    GPU,
                    ContextProjectionBackwardDataFunc);
REGISTER_TYPED_FUNC(ContextProjectionBackwardWeight,
                    GPU,
                    ContextProjectionBackwardWeightFunc);
#endif
}  // namespace paddle