deeplearning4j--deeplearning4j/libnd4j/include/ops/declarable/generic/reduce/reduce_norm1.cpp

/* ******************************************************************************
 *
 *
 * 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
 ******************************************************************************/

//
// Created by george@skymind.io on 6/4/2018.
// @author Yurii Shyrma (iuriish@yahoo.com)
//

#include <ops/declarable/CustomOperations.h>
#include <ops/declarable/helpers/axis.h>

namespace sd {
namespace ops {
#if NOT_EXCLUDED(OP_reduce_norm1)

//////////////////////////////////////////////////////////////////////////
CUSTOM_OP_IMPL(reduce_norm1, -1, 1, false, 0, 0) {
  auto input = INPUT_VARIABLE(0);
  auto output = OUTPUT_VARIABLE(0);

  std::vector<sd::LongType> dimensions;
  if (block.width() > 1) {
    auto axesVector = INPUT_VARIABLE(1);
    helpers::adjustAxis(input->rankOf(), axesVector, dimensions);
  } else if (block.getIArguments()->size())
    dimensions = *block.getIArguments();

  REQUIRE_TRUE(
      dimensions.size() <= static_cast<size_t>(input->rankOf()), 0,
      "REDUCE_NORM1 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->shapeInfo()[0] && item < input->shapeInfo()[0], 0,
        "REDUCE_NORM1 OP: the input dimension to reduce along must be in range [-%i, %i), but got %i instead !",
        input->rankOf(), input->rankOf(), item);

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

  input->reduceAlongDimension(reduce::Norm1, output, &dimensions, keepDims);

  return sd::Status::OK;
}

DECLARE_SHAPE_FN(reduce_norm1) {
  bool keepDims = false;
  if (block.getBArguments()->size())
    keepDims = B_ARG(0);
  else if (block.getTArguments()->size())
    keepDims = (bool)T_ARG(0);

  std::vector<sd::LongType> dimensions;
  if (block.width() > 1) {
    auto axesVector = INPUT_VARIABLE(1);
    helpers::adjustAxis(INPUT_VARIABLE(0)->rankOf(), axesVector, dimensions);
  } else if (block.getIArguments()->size())
    dimensions = *block.getIArguments();

  REQUIRE_TRUE(
      dimensions.size() <= static_cast<size_t>(inputShape->at(0)[0]), 0,
      "REDUCE_NORM1 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_NORM1 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);

  return SHAPELIST(ShapeUtils::evalReduceShapeInfo(shape::order(inputShape->at(0)), &dimensions, inputShape->at(0),
                                                   keepDims, false, block.getWorkspace()));
}

DECLARE_TYPES(reduce_norm1) {
  getOpDescriptor()->setAllowedInputTypes(sd::DataType::ANY)->setAllowedOutputTypes({ALL_FLOATS});
}
#endif
#if NOT_EXCLUDED(OP_reduce_norm1_bp)

//////////////////////////////////////////////////////////////////////////
CUSTOM_OP_IMPL(reduce_norm1_bp, -1, 1, false, 0, 0) {
  // L = Sum abs(x_i) for all i = 1 to N
  // dL/dx_i = 1 if x_i >= 0 and -1 when x_i < 0
  // out_i = epsilon_i if x_i > 0 and -epsilon_i when x_i < 0
  // when gradO is non a scalar, using dimensions to split output onto gradO like parts
  // and use LAMBDA with that formula for it.

  auto input = INPUT_VARIABLE(0);
  auto gradO = INPUT_VARIABLE(1);
  auto gradI = OUTPUT_VARIABLE(0);

  input->applyTransform(sd::transform::Sign, gradI);

  bool keepDims = false;
  auto dimensions = *block.getIArguments();

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

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

  REQUIRE_TRUE(
      dimensions.size() <= static_cast<size_t>(input->rankOf()), 0,
      "REDUCE_NORM1_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 >= -input->rankOf() && item < input->rankOf(), 0,
        "REDUCE_NORM1_BP OP: the input dimension to reduce along must be in range [-%i, %i), but got %i instead !",
        input->rankOf(), input->rankOf(), item);

  // *** calculations *** //

  if (!keepDims && gradO->lengthOf() > 1) {
    auto gradOShapeKeepDims =
        ShapeUtils::evalReduceShapeInfo(gradO->ordering(), &dimensions, *input, true, false, block.getWorkspace());
    std::vector<sd::LongType> shape =  ShapeUtils::pullShapeFromShapeInfo(
        gradOShapeKeepDims);
    auto reshaped = gradO->reshape(gradO->ordering(),
                                   shape);
    *gradI *= *reshaped;  // for example could be something like [a,b] -> [1,a,1,b]
    delete reshaped;
  } else
    *gradI *= *gradO;

  return sd::Status::OK;
}

DECLARE_SHAPE_FN(reduce_norm1_bp) {
  auto dimensions = *block.getIArguments();
  if (block.width() > 2) {
    auto axesVector = INPUT_VARIABLE(2);
    helpers::adjustAxis(INPUT_VARIABLE(0)->rankOf(), axesVector, dimensions);
  }

  REQUIRE_TRUE(
      dimensions.size() <= static_cast<size_t>(inputShape->at(0)[0]), 0,
      "REDUCE_NORM1_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 >= -inputShape->at(0)[0] && item < inputShape->at(0)[0], 0,
        "REDUCE_NORM1_BP 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);

  return SHAPELIST(CONSTANT(inputShape->at(0)));
}

DECLARE_TYPES(reduce_norm1_bp) {
  getOpDescriptor()->setAllowedInputTypes(sd::DataType::ANY)->setAllowedOutputTypes({ALL_FLOATS});
}

#endif

}  // namespace ops
}  // namespace sd