|
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158 |
- ///-------------------------------------------------------------------------------------------------
- ///
- /// \file test_Misc.hpp
- /// \brief Tests for Misc Functions of module.
- /// \author Thibaut Monseigne (Inria).
- /// \version 1.0.
- /// \date 29/07/2020.
- /// \copyright <a href="https://choosealicense.com/licenses/agpl-3.0/">GNU Affero General Public License v3.0</a>.
- /// \remarks We use the EEglab Matlab plugin to compare result for validation
- ///
- ///-------------------------------------------------------------------------------------------------
-
- #pragma once
-
- #include "gtest/gtest.h"
- #include "init.hpp"
- #include "misc.hpp"
-
- #include <geometry/Misc.hpp>
- #include <geometry/Basics.hpp>
-
- //---------------------------------------------------------------------------------------------------
- class Tests_Misc : public testing::Test
- {
- //protected:
- // std::vector<Eigen::MatrixXd> m_dataSet;
- //
- // void SetUp() override { m_dataSet = Vector2DTo1D(InitCovariance::LWF::Reference()); }
- };
-
- //---------------------------------------------------------------------------------------------------
-
- //---------------------------------------------------------------------------------------------------
- TEST_F(Tests_Misc, Double_Range)
- {
- const std::vector<double> calc1 = Geometry::doubleRange(0, 10, 2), calc2 = Geometry::doubleRange(0, 10, 2, false),
- calc3 = Geometry::doubleRange(0.15, 3.05, 0.5), calc4 = Geometry::doubleRange(0.15, 3.05, 0.5, false),
- ref1 = { 0, 2, 4, 6, 8, 10 }, ref2 = { 0, 2, 4, 6, 8 },
- ref3 = { 0.15, 0.65, 1.15, 1.65, 2.15, 2.65 }, ref4 = { 0.15, 0.65, 1.15, 1.65, 2.15, 2.65 };
-
- EXPECT_TRUE(isAlmostEqual(ref1, calc1)) << ErrorMsg("Double closed Range with integer value", ref1, calc1);
- EXPECT_TRUE(isAlmostEqual(ref2, calc2)) << ErrorMsg("Double opened Range with integer value", ref2, calc2);
- EXPECT_TRUE(isAlmostEqual(ref3, calc3)) << ErrorMsg("Double closed Range with double value", ref3, calc3);
- EXPECT_TRUE(isAlmostEqual(ref4, calc4)) << ErrorMsg("Double opened Range with double value", ref4, calc4);
- }
- //---------------------------------------------------------------------------------------------------
-
- //---------------------------------------------------------------------------------------------------
- TEST_F(Tests_Misc, Round_Index_Range)
- {
- const std::vector<size_t> calc1 = Geometry::RoundIndexRange(0, 10, 2), calc2 = Geometry::RoundIndexRange(0, 10, 2, false),
- calc3 = Geometry::RoundIndexRange(0.15, 3.15, 0.2), calc4 = Geometry::RoundIndexRange(0.15, 3.05, 0.2, false, false),
- ref1 = { 0, 2, 4, 6, 8, 10 }, ref2 = { 0, 2, 4, 6, 8 },
- ref3 = { 0, 1, 2, 3 }, ref4 = { 0, 0, 1, 1, 1, 1, 1, 2, 2, 2, 2, 2, 3, 3, 3 };
-
- EXPECT_TRUE(isAlmostEqual(ref1, calc1)) << ErrorMsg("Round Index closed Range with integer value", ref1, calc1);
- EXPECT_TRUE(isAlmostEqual(ref2, calc2)) << ErrorMsg("Round Index opened Range with integer value", ref2, calc2);
- EXPECT_TRUE(isAlmostEqual(ref3, calc3)) << ErrorMsg("Round Index closed Range with double value", ref3, calc3);
- EXPECT_TRUE(isAlmostEqual(ref4, calc4)) << ErrorMsg("Round Index opened Range with double value", ref4, calc4);
- }
- //---------------------------------------------------------------------------------------------------
-
- //---------------------------------------------------------------------------------------------------
- TEST_F(Tests_Misc, Bin_Histogramm)
- {
- //========== Create Dataset ==========
- const std::vector<Eigen::MatrixXd> matrices = Geometry::Vector2DTo1D(InitDataset::Dataset());
- std::vector<std::vector<double>> dataset(NB_CHAN);
- // Transform Dataset to vector per channel
- for (size_t i = 0; i < NB_CHAN; ++i) { dataset[i].reserve(NB_SAMPLE * matrices.size()); }
- for (const auto& m : matrices) { for (size_t i = 0; i < NB_CHAN; ++i) { for (size_t j = 0; j < NB_SAMPLE; ++j) { dataset[i].push_back(m(i, j)); } } }
-
- // Sort and remove first (to begin by 0)
- for (auto& d : dataset)
- {
- std::sort(d.begin(), d.end());
- const auto first = d[0];
- for (auto& e : d) { e -= first; }
- }
-
- //========== Create Ref ==========
- const std::vector<std::vector<size_t>> ref =
- {
- { 12, 10, 0, 15, 15, 0, 6, 12, 0, 11, 11, 0, 11, 14, 3 },
- { 17, 0, 26, 0, 0, 18, 0, 28, 0, 0, 15, 0, 7, 0, 9 },
- { 36, 0, 0, 34, 0, 0, 0, 0, 0, 15, 0, 0, 20, 0, 15 }
- };
-
- //========== Test ==========
- for (size_t i = 0; i < NB_CHAN; ++i)
- {
- const std::vector<size_t> hist = Geometry::BinHist(dataset[i], 15);
- EXPECT_TRUE(isAlmostEqual(hist, ref[i])) << ErrorMsg("Bin Histogramm", hist, ref[i]);
- }
- }
- //---------------------------------------------------------------------------------------------------
-
- //---------------------------------------------------------------------------------------------------
- TEST_F(Tests_Misc, Fit_Distribution)
- {
- const std::vector<Eigen::MatrixXd> matrices = Geometry::Vector2DTo1D(InitDataset::Dataset());
- std::vector<std::vector<double>> dataset(NB_CHAN);
- // Transform Dataset to vector per channel
- for (size_t i = 0; i < NB_CHAN; ++i) { dataset[i].reserve(NB_SAMPLE * matrices.size()); }
- for (const auto& m : matrices) { for (size_t i = 0; i < NB_CHAN; ++i) { for (size_t j = 0; j < NB_SAMPLE; ++j) { dataset[i].push_back(m(i, j)); } } }
-
- // Begin Fit Distribution
- std::vector<double> mu(NB_CHAN), sigma(NB_CHAN);
- const std::vector<double> refMu = { -0.840258269642149, - 2.10169835819046, 0.898301641809541 },
- refSigma = { 2.76541902273525, 0.435493584265319, 0.435493584265319 };
-
- for (size_t i = 0; i < NB_CHAN; ++i)
- {
- Geometry::FitDistribution(dataset[i], mu[i], sigma[i]);
- EXPECT_TRUE(isAlmostEqual(mu[i], refMu[i])) << ErrorMsg("Fit Distribution Mu", mu[i], refMu[i]);
- EXPECT_TRUE(isAlmostEqual(sigma[i], refSigma[i])) << ErrorMsg("Fit Distribution Sigma", sigma[i], refSigma[i]);
- }
- }
- //---------------------------------------------------------------------------------------------------
-
- //---------------------------------------------------------------------------------------------------
- TEST_F(Tests_Misc, Sorted_Eigen_Vector_Euclidian)
- {
- std::vector<Eigen::MatrixXd> matrices = Geometry::Vector2DTo1D(InitCovariance::LWF::Reference());
- const size_t n = matrices.size();
- std::vector<Eigen::MatrixXd> vectors = InitEigenVector::Euclidian::Vectors();
- std::vector<std::vector<double>> values = InitEigenVector::Euclidian::Values();
- for (size_t i = 0; i < n; ++i)
- {
- Eigen::MatrixXd vec;
- std::vector<double> val;
- Geometry::sortedEigenVector(matrices[i], vec, val, Geometry::EMetric::Euclidian);
- EXPECT_TRUE(isAlmostEqual(vectors[i], vec)) << ErrorMsg("Eigen Vector sample " + std::to_string(i) + " : ", vectors[i], vec);
- EXPECT_TRUE(isAlmostEqual(values[i], val)) << ErrorMsg("Eigen Value sample " + std::to_string(i) + " : ", values[i], val);
- }
- }
- //---------------------------------------------------------------------------------------------------
-
- /*
- //---------------------------------------------------------------------------------------------------
- TEST_F(Tests_Misc, Sorted_Eigen_Vector_Riemann)
- {
- std::cout << "Not implemented" << std::endl;
- std::vector<Eigen::MatrixXd> matrices = Geometry::Vector2DTo1D(InitCovariance::LWF::Reference());
- const size_t n = matrices.size();
- //std::vector<Eigen::MatrixXd> vectors = InitEigenVector::Riemann::Vectors();
- //std::vector<std::vector<double>> values = InitEigenVector::Riemann::Values();
- for (size_t i = 0; i < n; ++i)
- {
- Eigen::MatrixXd vec;
- std::vector<double> val;
- Geometry::sortedEigenVector(matrices[i], vec, val, Geometry::EMetric::Riemann);
- //EXPECT_TRUE(isAlmostEqual(vectors[i], vec)) << ErrorMsg("Eigen Vector sample " + std::to_string(i) + " : ", vectors[i], vec);
- //EXPECT_TRUE(isAlmostEqual(values[i], val)) << ErrorMsg("Eigen Value sample " + std::to_string(i) + " : ", values[i], val);
- }
- }
- //---------------------------------------------------------------------------------------------------
- */
|