metric.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_METRIC_HPP_INCLUDED
#define LBANN_METRIC_HPP_INCLUDED

#include "lbann/base.hpp"

#include <map>
#include <typeindex>
#include <vector>

namespace lbann {

// Forward declarations
class model;
class Layer;
class lbann_comm;
using ViewingLayerPtr = std::weak_ptr<Layer>;
class persist;

struct metric_statistics
{
  EvalType m_sum;
  int m_num_samples;
  metric_statistics() { reset(); }
  metric_statistics(metric_statistics& other) = default;
  metric_statistics(const metric_statistics& other) = default;
  metric_statistics& operator=(metric_statistics& other) = default;
  metric_statistics& operator=(const metric_statistics& other) = default;
  ~metric_statistics() = default;

  template <class Archive>
  void serialize(Archive& ar);

  void add_value(EvalType value, int num_samples = 1);
  EvalType get_mean() const;
  int get_num_samples() const { return m_num_samples; }
  void reset();
};

class metric
{

public:
  metric(lbann_comm* comm);

  metric(const metric& other) = default;
  metric& operator=(const metric& other) = default;
  virtual ~metric() = default;
  virtual metric* copy() const = 0;

  template <class Archive>
  void serialize(Archive& ar);

  virtual std::string name() const = 0;
  virtual std::string get_unit() const { return ""; }

  virtual void setup(model& m) {}

  virtual EvalType evaluate(execution_mode mode, int mini_batch_size) = 0;

  void reset_statistics()
  {
    for (auto& stats : m_statistics) {
      stats.second.reset();
    }
  }
  void reset_statistics(execution_mode mode) { m_statistics[mode].reset(); }

  EvalType get_mean_value(execution_mode mode) const;
  int get_statistics_num_samples(execution_mode mode) const;

  virtual std::vector<ViewingLayerPtr> get_layer_pointers() const;
  virtual void set_layer_pointers(std::vector<ViewingLayerPtr> layers);

  EvalType get_evaluate_time() const { return m_evaluate_time; }
  EvalType& get_evaluate_time() { return m_evaluate_time; }

  void reset_counters() { m_evaluate_time = 0.0; }

  virtual bool save_to_checkpoint_shared(persist& p) = 0;
  virtual bool load_from_checkpoint_shared(persist& p) = 0;

  virtual bool save_to_checkpoint_distributed(persist& p) = 0;
  virtual bool load_from_checkpoint_distributed(persist& p) = 0;

protected:
  virtual EvalType evaluate_compute(const AbsDistMat& prediction,
                                    const AbsDistMat& ground_truth) = 0;

  lbann_comm& get_comm() { return *m_comm; }

  std::map<execution_mode, metric_statistics>& get_statistics()
  {
    return m_statistics;
  }

private:
  lbann_comm* m_comm;

  std::map<execution_mode, metric_statistics> m_statistics;

  EvalType m_evaluate_time = 0.0;
};

} // namespace lbann

#endif // LBANN_METRIC_HPP_INCLUDED