public void Test_01()
{
// ???
Matrix data = new Matrix(
new RowVector(86, 67),
new RowVector(71, 78),
new RowVector(42, 39),
new RowVector(62, 98),
new RowVector(96, 61),
new RowVector(39, 45),
new RowVector(50, 64),
new RowVector(78, 52),
new RowVector(51, 76),
new RowVector(89, 93));
Pca result = Func.Pca(data, cor: true);
Assert.That(result.Ratios[0], Is.EqualTo(0.6880464).Within(Ep));
Assert.That(result.Ratios[1], Is.EqualTo(0.3119536).Within(Ep));
var ir2 = Math.Sqrt(2);
Func <double, double> abs = v => Math.Abs(v);
// 正負の自由があるため，π の差異は同じとみなす
Assert.That(abs(result.Eigenvectors[0][0]), Is.EqualTo(1 / ir2).Within(Ep));
Assert.That(abs(result.Eigenvectors[0][1]), Is.EqualTo(1 / ir2).Within(Ep));
Assert.True(result.Eigenvectors[0][0] * result.Eigenvectors[0][1] > 0);
Assert.That(abs(result.Eigenvectors[1][0]), Is.EqualTo(1 / ir2).Within(Ep));
Assert.That(abs(result.Eigenvectors[1][1]), Is.EqualTo(1 / ir2).Within(Ep));
Assert.True(result.Eigenvectors[1][0] * result.Eigenvectors[1][1] < 0);
}
public void Test_02()
{
// サイエンス社 ライブラリ新数学大系 E20 「多変量解析法入門」 (初版第 7 刷発行)
// 永田 靖・棟近雅彦 共著
// P.132 表 9.1
var data = new Matrix(
new[, ] {
{ 86.0, 79, 67, 68 },
{ 71, 75, 78, 84 },
{ 42, 43, 39, 44 },
{ 62, 58, 98, 95 },
{ 96, 97, 61, 63 },
{ 39, 33, 45, 50 },
{ 50, 53, 64, 72 },
{ 78, 66, 52, 47 },
{ 51, 44, 76, 72 },
{ 89, 92, 93, 91 }
});
Pca res = Func.Pca(data);
Assert.That(res.Ratios[0], Is.EqualTo(0.6889703).Within(Ep));
Assert.That(res.Ratios[1], Is.EqualTo(0.2965368).Within(Ep));
Assert.That(res.Ratios[2], Is.EqualTo(0.01326883).Within(Ep));
Assert.That(res.Ratios[3], Is.EqualTo(0.001224026).Within(Ep));
Func <double, double> abs = v => Math.Abs(v);
/* R の結果 (分散共分散行列)
* Comp.1 Comp.2 Comp.3 Comp.4
* 国語 -0.545 0.440 -0.549 0.456
* 英語 -0.592 0.401 0.556 -0.424
* 数学 -0.443 -0.576 -0.452 -0.518
* 理科 -0.395 -0.561 0.431 0.586
*/
var vecs = res.Eigenvectors;
Assert.That(abs(vecs[0][0]), Is.EqualTo(abs(-0.545)).Within(Ep2));
Assert.That(abs(vecs[0][1]), Is.EqualTo(abs(-0.592)).Within(Ep2));
Assert.That(abs(vecs[0][2]), Is.EqualTo(abs(-0.443)).Within(Ep2));
Assert.That(abs(vecs[0][3]), Is.EqualTo(abs(-0.395)).Within(Ep2));
Assert.True(vecs[0][0] * vecs[0][1] * vecs[0][2] * vecs[0][3] > 0);
Assert.That(abs(vecs[1][0]), Is.EqualTo(abs(0.440)).Within(Ep2));
Assert.That(abs(vecs[1][1]), Is.EqualTo(abs(0.401)).Within(Ep2));
Assert.That(abs(vecs[1][2]), Is.EqualTo(abs(-0.576)).Within(Ep2));
Assert.That(abs(vecs[1][3]), Is.EqualTo(abs(-0.561)).Within(Ep2));
Assert.True(Math.Sign(vecs[1][0]) == Math.Sign(vecs[1][1]));
Assert.True(Math.Sign(vecs[1][2]) == Math.Sign(vecs[1][3]));
Assert.False(Math.Sign(vecs[1][1]) == Math.Sign(vecs[1][2]));
}
public void TestKnn()
{
var knn = new Knn();
var knnParameter = new KnnParameter(1, 3, new EuclidianDistance());
var pca = new Pca(iris);
pca.Convert();
knn.Train(iris.GetInstanceList(), knnParameter);
Assert.AreEqual(4.00, 100 * knn.Test(iris.GetInstanceList()).GetErrorRate(), 0.01);
pca = new Pca(bupa);
pca.Convert();
knn.Train(bupa.GetInstanceList(), knnParameter);
Assert.AreEqual(19.13, 100 * knn.Test(bupa.GetInstanceList()).GetErrorRate(), 0.01);
pca = new Pca(dermatology);
pca.Convert();
knn.Train(dermatology.GetInstanceList(), knnParameter);
Assert.AreEqual(3.28, 100 * knn.Test(dermatology.GetInstanceList()).GetErrorRate(), 0.01);
}
public void TestLinearPerceptron()
{
var linearPerceptron = new LinearPerceptron();
var linearPerceptronParameter = new LinearPerceptronParameter(1, 0.1, 0.99, 0.2, 100);
var pca = new Pca(iris);
pca.Convert();
linearPerceptron.Train(iris.GetInstanceList(), linearPerceptronParameter);
Assert.AreEqual(8.67, 100 * linearPerceptron.Test(iris.GetInstanceList()).GetErrorRate(), 0.01);
linearPerceptronParameter = new LinearPerceptronParameter(1, 0.01, 0.99, 0.2, 100);
pca = new Pca(bupa);
pca.Convert();
linearPerceptron.Train(bupa.GetInstanceList(), linearPerceptronParameter);
Assert.AreEqual(42.03, 100 * linearPerceptron.Test(bupa.GetInstanceList()).GetErrorRate(), 0.01);
pca = new Pca(dermatology);
pca.Convert();
linearPerceptron.Train(dermatology.GetInstanceList(), linearPerceptronParameter);
Assert.AreEqual(1.64, 100 * linearPerceptron.Test(dermatology.GetInstanceList()).GetErrorRate(), 0.01);
}