metrics.h File Reference

#include "kernel/gp/src/predictor.h"
#include <cmath>
#include <limits>
#include <ranges>

Go to the source code of this file.

Namespaces
namespace	ultra

Functions
template<ClassificationPredictor P>
double	ultra::src::metrics::accuracy (const P &p, const dataframe &d)
template<RegressionPredictor P>
double	ultra::src::metrics::mae (const P &p, const dataframe &d)

Detailed Description

Remarks

This file is part of ULTRA.

Copyright

Copyright (C) 2026 EOS di Manlio Morini.

License

This Source Code Form is subject to the terms of the Mozilla Public License, v. 2.0. If a copy of the MPL was not distributed with this file, You can obtain one at http://mozilla.org/MPL/2.0/

Function Documentation

accuracy()

template<ClassificationPredictor P>

double ultra::src::metrics::accuracy ( const P & p, const dataframe & d )

nodiscard

Accuracy for classification problems.

Parameters

[in]	p	oracle-like object
[in]	d	classification dataset

Returns

fraction of correctly classified examples

Accuracy refers to the number of training examples that are correctly valued/classified as a proportion of the total number of examples in the training set.

According to this design, the best accuracy is 1.0 (100%), meaning that all the training examples have been correctly recognized.

Note

Accuracy and fitness aren't the same thing. Accuracy can be used to measure fitness but often it hasn't enough "granularity"; also it isn't appropriated for classification tasks with imbalanced learning data (where at least one class is under/over represented relative to others).

mae()

template<RegressionPredictor P>

double ultra::src::metrics::mae ( const P & p, const dataframe & d )

nodiscard

Mean absolute error for regression problems.

Missing predictions are ignored. If no valid prediction is produced, the function returns infinity().

Parameters

[in]	p	oracle-like object
[in]	d	regression dataset

Returns

mean absolute error over valid predictions