training_result.h

Go to the documentation of this file.

// SPDX-License-Identifier: AGPL-3.0-only
// Copyright (C) 2026 Artificial Intelligence Techniques, S.L. and Roberto López González
 
#pragma once
 
#include "opennn/core/opennn_types.h"
#include "opennn/core/tensor_types.h"
 
namespace opennn
{
 
// Listed in reporting priority: convergence, early stopping, then the budgets.
enum class StoppingCondition {LossGoal,
                              MinimumLossDecrease,
                              MaximumValidationErrorIncreases,
                              MaximumTime,
                              MaximumEpochsNumber};
 
struct TrainingResult
{
    TrainingResult(const Index = 0);
    virtual ~TrainingResult() = default;
 
    string write_stopping_condition() const;
 
    // Report the restored model's epoch when available; histories retain every epoch.
    float get_training_error() const;
 
    // Without restoration, return the latest finite measurement, or NaN if unavailable.
    float get_validation_error() const;
 
    Index get_epochs_number() const { return training_error_history.size(); }
 
    void save(const filesystem::path&) const;
 
    void print(const string& message = {}) const;
 
    optional<StoppingCondition> stopping_condition;
 
    Tensor<string, 2> write_override_results(const Index = 3) const;
 
    void resize_training_error_history(const Index new_size) { training_error_history.conservativeResize(new_size); }
 
    void resize_validation_error_history(const Index new_size) { validation_error_history.conservativeResize(new_size); }
 
    VectorR training_error_history;
 
    VectorR validation_error_history;
 
    string elapsed_time;
 
    double training_seconds = 0.0;
 
    float loss = QUIET_NAN;
 
    optional<Index> restored_epoch;
};
 
}