SelectiveFullyConnectedLayer.cpp 12.1 KB
Newer Older
1
/* Copyright (c) 2016 PaddlePaddle Authors. All Rights Reserve.
Z
zhangjinchao01 已提交
2 3 4 5 6 7 8 9 10 11 12 13 14 15

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 "SelectiveFullyConnectedLayer.h"
Y
Yu Yang 已提交
16 17 18
#include <algorithm>
#include <vector>
#include "paddle/math/SparseMatrix.h"
Z
zhangjinchao01 已提交
19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
#include "paddle/utils/Logging.h"
#include "paddle/utils/Stat.h"

namespace paddle {

REGISTER_LAYER(selective_fc, SelectiveFullyConnectedLayer);

bool SelectiveFullyConnectedLayer::init(const LayerMap& layerMap,
                                        const ParameterMap& parameterMap) {
  Layer::init(layerMap, parameterMap);
  inputNum_ = inputLayers_.size();
  if (config_.has_selected_colums()) {
    inputNum_ -= 1;
  }
  for (size_t i = 0; i < inputNum_; i++) {
    size_t height = inputLayers_[i]->getSize();
    size_t width = getSize();
    // NOTE weight is transpoed
    weights_.emplace_back(new Weight(width, height, parameters_[i]));
  }

  if (biasParameter_.get() != NULL) {
    biases_ = std::unique_ptr<Weight>(new Weight(1, getSize(), biasParameter_));
  }

  fullOutput_ = false;

  return true;
}

void SelectiveFullyConnectedLayer::prefetch() {}

51 52
void SelectiveFullyConnectedLayer::reserveOutput(size_t height,
                                                 size_t width,
Z
zhangjinchao01 已提交
53 54
                                                 size_t nnz) {
  bool flag = (passType_ == PASS_TEST &&
55
               config_.selective_fc_pass_generation() && !fullOutput_);
Z
zhangjinchao01 已提交
56 57 58 59 60 61 62
  SetDevice device(output_.deviceId);
  if (flag) {
    // output_.value is sparse matrix
    if (dynamic_cast<CpuMatrix*>(output_.value.get()) ||
        dynamic_cast<GpuMatrix*>(output_.value.get())) {
      output_.value = nullptr;
    }
63 64 65 66 67 68
    Matrix::resizeOrCreateSparseMatrix(output_.value,
                                       height,
                                       width,
                                       nnz,
                                       FLOAT_VALUE,
                                       SPARSE_CSR,
Z
zhangjinchao01 已提交
69 70 71 72 73 74 75 76 77 78 79
                                       /*trans=*/false,
                                       /*useGpu=*/useGpu_);
    output_.value->copyFrom(*selCols_);
    interOutput_ = output_.value;
  } else {
    if (fullOutput_) {
      // output_.value is dense matrix
      if (dynamic_cast<CpuSparseMatrix*>(output_.value.get()) ||
          dynamic_cast<GpuSparseMatrix*>(output_.value.get())) {
        output_.value = nullptr;
      }
80 81 82 83 84
      Matrix::resizeOrCreate(output_.value,
                             height,
                             width,
                             /*trans=*/false,
                             /*useGpu=*/useGpu_);
Z
zhangjinchao01 已提交
85 86 87 88 89
      interOutput_ = output_.value;
    } else {
      // output_.value is dense matrix, but width = nnz /height
      CHECK_EQ(nnz % height, 0U);
      CHECK(nnz / height);
90 91 92 93 94 95 96 97 98 99 100 101 102 103 104
      Matrix::resizeOrCreate(output_.value,
                             height,
                             nnz / height,
                             /*trans=*/false,
                             /*useGpu=*/useGpu_);
      interOutput_ = Matrix::createSparseMatrix(output_.value->getData(),
                                                selCols_->getRows(),
                                                selCols_->getCols(),
                                                height,
                                                width,
                                                nnz,
                                                FLOAT_VALUE,
                                                SPARSE_CSR,
                                                /*trans=*/false,
                                                /*useGpu=*/useGpu_);
Z
zhangjinchao01 已提交
105 106 107 108 109 110 111 112 113 114
    }
  }
  interOutput_->zeroMem();

  if (passType_ != PASS_TEST && needGradient()) {
    CHECK_EQ(nnz % height, 0U) << "during training, each sample must have a "
                                  "same number of selected columns.";
    CHECK(nnz / height)
        << "during training, "
           "each sample must have at least one column selected.";
115 116 117 118 119
    Matrix::resizeOrCreate(output_.grad,
                           height,
                           nnz / height,
                           /*trans=*/false,
                           /*useGpu=*/useGpu_);
Z
zhangjinchao01 已提交
120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151
    output_.grad->zeroMem();
  }
}

void SelectiveFullyConnectedLayer::forward(PassType passType) {
  REGISTER_TIMER("selective_fc.forward");
  Layer::forward(passType);

  getSelectiveCols();
  size_t height = getInput(0).getBatchSize();
  size_t width = getSize();
  size_t nnz = height * width;
  if (!fullOutput_) {
    CHECK(selCols_);
    CHECK(height == selCols_->getHeight());
    CHECK(width == selCols_->getWidth());
    nnz = selCols_->getElementCnt();
  }

  // Layer::ResetOutput(), here we set outV/outG as SparseMatrix manually
  // this outV should be used as input of MaxIdLayer and softmax activation
  reserveOutput(height, width, nnz);

  bool flag = true;
  for (size_t i = 0; i < inputNum_; i++) {
    MatrixPtr input = getInputValue(i);
    MatrixPtr weight = weights_[i]->getW();
    size_t hsize = input->getHeight();
    size_t wsize = weight->getHeight();
    real scaleT = i == 0 ? real(0) : real(1);

    flag = nnz < (hsize * wsize) * config_.selective_fc_full_mul_ratio() &&
152
           !fullOutput_;
Z
zhangjinchao01 已提交
153 154 155 156 157 158 159 160 161 162 163 164 165
    if (flag) {
      // if the indecies are highly sparse,
      // manully compute the multiplication of
      // the input vector and the selected rows.
      REGISTER_TIMER("selective.plain");
      interOutput_->mul(input, weight->getTranspose(), 1, scaleT);
    } else {
      // if the indecies is not sparse enough,
      // use full mul instead
      REGISTER_TIMER("selective.mul");
      if (fullOutput_) {
        interOutput_->mul(input, weight->getTranspose(), 1, scaleT);
      } else {
166 167 168 169 170
        Matrix::resizeOrCreate(mmat_,
                               hsize,
                               wsize,
                               /*trans=*/false,
                               /*useGpu=*/useGpu_);
Z
zhangjinchao01 已提交
171 172 173 174 175 176 177 178 179 180 181
        mmat_->mul(input, weight->getTranspose());
        interOutput_->add3(mmat_);
      }
    }
  }

  if (biases_) {
    interOutput_->addBias(*(biases_->getW()), 1);
  }

  flag = (passType_ == PASS_TEST && config_.selective_fc_pass_generation() &&
182
          !fullOutput_);
Z
zhangjinchao01 已提交
183 184 185 186 187 188 189
  if (flag) {
    // during generation, output of this layer is a sparse csr matrix,
    // which is probably the input of maxid layer
    // if the model is trained with multi-class-cross-entroy-with-selfnorm,
    // activiation of this layer should be exponential, not softmax.

    Argument arg;
190 191 192 193 194
    arg.value = Matrix::create(interOutput_->getData(),
                               1,
                               nnz,
                               /*trans=*/false,
                               /*useGpu=*/useGpu_);
Z
zhangjinchao01 已提交
195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213
    activation_->forward(arg);
  } else /* train and test in train, not generating */ {
    // during training, this layer output value is *Matrix*, which is input of
    // eg. multi-class-cross-entropy

    // while training, every sample has a equal number of selected
    // columns to be activated.
    // note indices of multi-class-cross-entropy need to be remapped
    // to this index.
    // e.g. sample = [1,3,5] and 3 is gold, then label is 1

    forwardActivation();
  }
}

void SelectiveFullyConnectedLayer::backward(const UpdateCallback& callback) {
  backwardActivation();
  MatrixPtr oGrad = getOutputGrad();
  if (!fullOutput_) {
214 215 216 217 218 219 220 221 222 223
    interOutGrad_ = Matrix::createSparseMatrix(oGrad->getData(),
                                               interOutput_->getRows(),
                                               interOutput_->getCols(),
                                               interOutput_->getHeight(),
                                               interOutput_->getWidth(),
                                               interOutput_->getElementCnt(),
                                               FLOAT_VALUE,
                                               SPARSE_CSR,
                                               /*trans=*/false,
                                               /*useGpu=*/useGpu_);
Z
zhangjinchao01 已提交
224
  } else {
225 226 227 228 229
    interOutGrad_ = Matrix::create(oGrad->getData(),
                                   oGrad->getHeight(),
                                   oGrad->getWidth(),
                                   /*trans=*/false,
                                   /*useGpu=*/useGpu_);
Z
zhangjinchao01 已提交
230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271
  }

  if (biases_ && biases_->getWGrad()) {
    REGISTER_TIMER_INFO("BpBiasTimer", getName().c_str());
    biases_->getWGrad()->collectBias(*interOutGrad_, 1);
    biases_->getParameterPtr()->incUpdate(callback);
  }

  // backward is different from FullyConnectedLayer
  // because the weight is transposed
  for (size_t i = 0; i < inputNum_; i++) {
    AsyncGpuBlock block;
    MatrixPtr preGrad = getInputGrad(i);
    if (preGrad) {
      REGISTER_TIMER_INFO("BpMulTimer", getName().c_str());
      preGrad->mul(interOutGrad_, weights_[i]->getW(), 1, 1);
    }

    MatrixPtr wGrad = weights_[i]->getWGrad();
    if (wGrad) {
      REGISTER_TIMER_INFO("GradMulTimer", getName().c_str());
      MatrixPtr input = getInputValue(i);
      wGrad->mul(interOutGrad_->getTranspose(), input, 1, 1);
    }

    {
      REGISTER_TIMER_INFO("WeightUpdate", getName().c_str());
      weights_[i]->getParameterPtr()->incUpdate(callback);
    }
  }
}

void paddle::SelectiveFullyConnectedLayer::fillSelectiveData(
    const std::shared_ptr<std::vector<std::pair<int*, size_t>>>& candidates) {
  if (candidates == nullptr) {
    fillFullySelectiveData();
    return;
  }

  size_t sampleNum = candidates->size();
  size_t outputWidth = getSize();
  size_t nnz =
272 273 274
      std::accumulate(candidates->begin(),
                      candidates->end(),
                      0UL,
Z
zhangjinchao01 已提交
275 276 277 278 279
                      [](size_t a, const std::pair<int*, size_t>& arr) {
                        return a + arr.second;
                      });

  Matrix::resizeOrCreateSparseMatrix(this->cpuSelCols_,
280 281 282 283 284 285 286
                                     sampleNum,
                                     outputWidth,
                                     nnz,
                                     NO_VALUE,
                                     SPARSE_CSR,
                                     false,
                                     false);
Z
zhangjinchao01 已提交
287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311
  CHECK(this->cpuSelCols_ != nullptr);
  CpuSparseMatrixPtr selCols =
      std::dynamic_pointer_cast<CpuSparseMatrix>(cpuSelCols_);
  int* rowOffsets = selCols->getRows();
  int* colIndices = selCols->getCols();

  rowOffsets[0] = 0;
  int idx = 0;
  for (size_t i = 0; i < sampleNum; ++i) {
    if ((*candidates)[i].second > 0) {
      rowOffsets[i + 1] = rowOffsets[i] + (*candidates)[i].second;
      for (size_t j = 0; j < (*candidates)[i].second; ++j) {
        colIndices[idx] = (*candidates)[i].first[j];
        idx++;
      }
    } else {
      rowOffsets[i + 1] = rowOffsets[i];
    }
  }

  CHECK_EQ(static_cast<size_t>(rowOffsets[sampleNum]), nnz);
  if (!useGpu_) {
    this->selCols_ = this->cpuSelCols_;
  } else {
    Matrix::resizeOrCreateSparseMatrix(this->selCols_,
312 313 314 315 316 317 318
                                       sampleNum,
                                       outputWidth,
                                       nnz,
                                       NO_VALUE,
                                       SPARSE_CSR,
                                       false,
                                       true);
Z
zhangjinchao01 已提交
319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335
    this->selCols_->copyFrom(*cpuSelCols_, HPPL_STREAM_1);
    hl_stream_synchronize(HPPL_STREAM_1);
  }

  fullOutput_ = false;
}

void paddle::SelectiveFullyConnectedLayer::getSelectiveCols() {
  if (config_.has_selected_colums()) {
    this->selCols_ = inputLayers_[inputNum_]->getOutputValue();
    fullOutput_ = false;
  } else if (!config_.selective_fc_pass_generation() || selCols_ == nullptr) {
    this->fillFullySelectiveData();
  }  // else selCols_ is initialized by fillSelectiveData
}

}  // namespace paddle