deeplearning4j/deeplearning4j

/* ******************************************************************************
 *
 *
 * This program and the accompanying materials are made available under the
 * terms of the Apache License, Version 2.0 which is available at
 * https://www.apache.org/licenses/LICENSE-2.0.
 *
 *  See the NOTICE file distributed with this work for additional
 *  information regarding copyright ownership.
 * 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.
 *
 * SPDX-License-Identifier: Apache-2.0
 ******************************************************************************/

//
// @author Yurii Shyrma (iuriish@yahoo.com), created on 01.06.2018
//
#include <ops/declarable/CustomOperations.h>
#include <system/op_boilerplate.h>

#if NOT_EXCLUDED(OP_reduce_mean)
#include <ops/declarable/helpers/axis.h>
namespace sd {
namespace ops {

//////////////////////////////////////////////////////////////////////////
CUSTOM_OP_IMPL(reduce_mean, -1, 1, false, 0, 0) {
  auto input = INPUT_VARIABLE(0);
  auto output = OUTPUT_VARIABLE(0);
  auto dimensions = *block.getIArguments();
  if (block.width() > 1) {
    auto axesVector = INPUT_VARIABLE(1);
    helpers::adjustAxis(input->rankOf(), axesVector, dimensions);
  }

  bool keepDims = false;
  if (block.getBArguments()->size())
    keepDims = B_ARG(0);
  else if (block.getTArguments()->size())
    keepDims = (bool)T_ARG(0);

  REQUIRE_TRUE(
      dimensions.size() <= input->rankOf(), 0,
      "REDUCE_MEAN OP: the number of dimensions to reduce along must be <= input array rank, but got %i instead",
      dimensions.size());

  for (const auto &item : dimensions) {
    REQUIRE_TRUE(item >= -input->rankOf() && item < input->rankOf(), 0,
                 "REDUCE_MEAN OP: the input dimension to reduce along must be in range [-%i, %i), but got %i instead !",
                 input->rankOf(), input->rankOf(), item);
  }

  input->reduceAlongDimension(reduce::Mean, *output, &dimensions, keepDims);
  return sd::Status::OK;
}

DECLARE_SHAPE_FN(reduce_mean) {
  auto dimensions = *block.getIArguments();
  auto in = inputShape->at(0);
  if (block.width() > 1) {
    auto axesVector = INPUT_VARIABLE(1);
    helpers::adjustAxis(shape::rank(in), axesVector, dimensions);
  }

  bool keepDims = false;
  if (block.getBArguments()->size())
    keepDims = B_ARG(0);
  else if (block.getTArguments()->size())
    keepDims = (bool)T_ARG(0);

  REQUIRE_TRUE(
      dimensions.size() <= in[0], 0,
      "REDUCE_MEAN OP: the number of dimensions to reduce along must be <= input array rank, but got %i instead",
      dimensions.size());

  for (const auto &item : dimensions)
  REQUIRE_TRUE(item >= -inputShape->at(0)[0] && item < inputShape->at(0)[0], 0,
               "REDUCE_MEAN OP: the input dimension to reduce along must be in range [-%i, %i), but got %i instead !",
               inputShape->at(0)[0], inputShape->at(0)[0], item);

  auto outShapeInfo =
      ShapeUtils::evalReduceShapeInfo(shape::order(in), &dimensions, in, keepDims, false, block.getWorkspace());

  return SHAPELIST(outShapeInfo);
}

DECLARE_TYPES(reduce_mean) {
  getOpDescriptor()->setAllowedInputTypes(sd::DataType::ANY)->setAllowedOutputTypes({ALL_FLOATS});
}

//////////////////////////////////////////////////////////////////////////
CUSTOM_OP_IMPL(reduce_mean_bp, -2, 1, false, 0, 0) {
  auto input = INPUT_VARIABLE(0);
  auto gradO = INPUT_VARIABLE(1);

  auto gradI = OUTPUT_VARIABLE(0);

  auto dimensions = *block.getIArguments();
  if (block.width() > 2) {
    auto axesVector = INPUT_VARIABLE(2);
    helpers::adjustAxis(input->rankOf(), axesVector, dimensions);
  }

  bool keepDims = false;
  if (block.getBArguments()->size())
    keepDims = B_ARG(0);
  else if (block.getTArguments()->size())
    keepDims = (bool)T_ARG(0);

  REQUIRE_TRUE(
      dimensions.size() <= input->rankOf(), 0,
      "REDUCE_MEAN_BP OP: the number of dimensions to reduce along must be <= input array rank, but got %i instead",
      dimensions.size());

  auto dimLength = 1.0;
  for (const auto &item : dimensions) {
    REQUIRE_TRUE(
        item >= -input->rankOf() && item < input->rankOf(), 0,
        "REDUCE_MEAN_BP OP: the input dimension to reduce along must be in range [-%i, %i), but got %i instead !",
        input->rankOf(), input->rankOf(), item);
    dimLength *= input->sizeAt(item);
  }
  if (gradO->isScalar()) {
    if (dimensions.size() > 0) {
      gradI->assign(gradO->e(0) / (static_cast<double>(dimLength)));
    } else {
      gradI->assign(gradO->e(0) / (static_cast<double>(input->lengthOf())));
    }
  } else {
    auto val = (static_cast<double>(gradO->lengthOf() < 1 ? 1.0 : gradO->lengthOf()) )
               / (static_cast<double>(input->lengthOf() < 1 ? 1.0 : input->lengthOf()));
    if(val == 0.0)
      val = SD_EPSILON;
    gradI->assign(val);
    if (!keepDims) {
      auto gradOShapeKeepDims =
          ShapeUtils::evalReduceShapeInfo(gradO->ordering(), &dimensions, *input, true, false, block.getWorkspace());
      *gradI *= gradO->reshape(gradO->ordering(),
                               ShapeUtils::pullShapeFromShapeInfo(
                                   gradOShapeKeepDims));  // for example could be something like [a,b] -> [1,a,1,b]
    } else {
      gradI->applyTrueBroadcast(sd::BroadcastOpsTuple::Multiply(), *gradO, *gradI);
    }
  }

  return sd::Status::OK;
}

DECLARE_SHAPE_FN(reduce_mean_bp) {
  auto in = inputShape->at(0);
  auto dimensions = *block.getIArguments();
  auto rank = shape::rank(in);

  if (block.width() > 2) {
    auto axesVector = INPUT_VARIABLE(2);
    helpers::adjustAxis(rank, axesVector, dimensions);
  }
  REQUIRE_TRUE(
      dimensions.size() <= rank, 0,
      "REDUCE_MEAN_BP OP: the number of dimensions to reduce along must be <= input array rank, but got %i instead",
      dimensions.size());

  for (const auto &item : dimensions)
  REQUIRE_TRUE(
      item >= -rank || item < rank, 0,
      "REDUCE_MEAN_BP OP: the input dimension to reduce along must be in range [-%i, %i), but got %i instead !", rank,
      rank, item);

  sd::LongType *gradIshapeInfo(nullptr);
  COPY_SHAPE(inputShape->at(0), gradIshapeInfo);
  return SHAPELIST(CONSTANT(gradIshapeInfo));
}

DECLARE_TYPES(reduce_mean_bp) {
  getOpDescriptor()->setAllowedInputTypes(sd::DataType::ANY)->setAllowedOutputTypes({ALL_FLOATS});
}

}  // namespace ops
}  // namespace sd
#endif