l2.hpp

Go to the documentation of this file.

// Copyright (c) 2014-2023, Lawrence Livermore National Security, LLC.
// Produced at the Lawrence Livermore National Laboratory.
// Written by the LBANN Research Team (B. Van Essen, et al.) listed in
// the CONTRIBUTORS file. <lbann-dev@llnl.gov>
//
// LLNL-CODE-697807.
// All rights reserved.
//
// This file is part of LBANN: Livermore Big Artificial Neural Network
// Toolkit. For details, see http://software.llnl.gov/LBANN or
// https://github.com/LLNL/LBANN.
//
// Licensed under the Apache License, Version 2.0 (the "Licensee"); 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.

#ifndef LBANN_OBJECTIVE_FUNCTIONS_WEIGHT_REGULARIZATION_L2_WEIGHT_REGULARIZATION_HPP_INCLUDED
#define LBANN_OBJECTIVE_FUNCTIONS_WEIGHT_REGULARIZATION_L2_WEIGHT_REGULARIZATION_HPP_INCLUDED

#include "lbann/comm_nb_request.hpp"
#include "lbann/objective_functions/objective_function_term.hpp"
#ifdef LBANN_HAS_GPU
#include "lbann/utils/gpu/helpers.hpp"
#endif // LBANN_HAS_GPU

namespace lbann {

template <typename>
class data_type_optimizer;
template <typename>
class data_type_weights;

class l2_weight_regularization : public objective_function_term
{
public:
  using AccumulateDataType = DataType;

  using OptimizerType = data_type_optimizer<DataType>;

  using WeightsType = data_type_weights<DataType>;

  template <El::Device D>
  using DMatType = El::Matrix<AccumulateDataType, D>;

  using CPUMatType = DMatType<El::Device::CPU>;

public:
  l2_weight_regularization(EvalType scale_factor = 1);
  l2_weight_regularization* copy() const override
  {
    return new l2_weight_regularization(*this);
  }

  template <typename ArchiveT>
  void serialize(ArchiveT& ar);

  std::string name() const override { return "L2 weight regularization"; }
  void setup(model& m) override;
  void start_evaluation() override;
  EvalType finish_evaluation() override;

  void differentiate() override{};

  void compute_weight_regularization() override;

private:
  void write_specific_proto(lbann_data::ObjectiveFunction& proto) const final;

  std::map<El::Device, CPUMatType> m_contributions;

  Al::request m_allreduce_req;
#ifdef LBANN_HAS_GPU

  gpu_lib::event_wrapper m_copy_event;
#endif // LBANN_HAS_GPU

  template <El::Device Device>
  static void accumulate_contribution(const DMatType<Device>& vals,
                                      DMatType<Device>& contribution);
};

} // namespace lbann

#endif // LBANN_OBJECTIVE_FUNCTIONS_WEIGHT_REGULARIZATION_L2_WEIGHT_REGULARIZATION_HPP_INCLUDED