0

commons-math3 和 ojalgo 库之间的 SVD 差异解决方案非常高。有什么方法可以减少基于任何输入参数的差异。

        double[][] olsColumns = { { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 },
                { 1.0, 1.0 }, { 1.0, 1.0 }, { 1.0, 1.0 } };
        double[] observationVector = { 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
                0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0 };

//Ojalgo
final PrimitiveDenseStore tmpOriginal = PrimitiveDenseStore.FACTORY.rows(olsColumns);
        SingularValue<Double> tmpSVD = SingularValue.make(tmpOriginal);
        tmpSVD.decompose(tmpOriginal);
        double[] singularValues = tmpSVD.getSingularValues().toRawCopy1D();
        double[][] V = tmpSVD.getQ2().toRawCopy2D();
        System.out.println("V" + Arrays.deepToString(V));
        System.out.println("Singular values" + Arrays.toString(singularValues));
        try {

            // MatrixStore<Double> doubleMat = tmpSVD.solve(tmpOriginal,
            // PrimitiveDenseStore.FACTORY.column(Utils.prepareObservationVector()));
            MatrixStore<Double> solution = tmpSVD.getSolution(PrimitiveDenseStore.FACTORY.column(observationVector),
                    tmpSVD.preallocate(tmpOriginal));
            System.out.println("Solution " + Arrays.toString(solution.toRawCopy1D()));
        } catch (Exception e) {
            e.printStackTrace();
        }

//Commons-Math3

        RealMatrix newPredM = new Array2DRowRealMatrix(olsColumns);
        SingularValueDecomposition svd = new SingularValueDecomposition(newPredM);
        // RealMatrix covariance = svd.getCovariance(0);
        // System.out.println("covariance"+Arrays.deepToString(covariance.getData()));
        System.out.println("V" + Arrays.deepToString(svd.getV().getData()));
        System.out.println("Singular values" + Arrays.toString(svd.getSingularValues()));
        double[] solution = svd.getSolver().solve(new ArrayRealVector(observationVector)).toArray();
        System.out.println("Solution" + Arrays.toString(solution));

Commons Math3 解决方案：[0.01612903225806451, 0.016129032258064502]

OjAlgo 解决方案解决方案：[7.614155324982286E13, -7.614155324982295E13]

4

1 回答 1

0

您使用的是哪个版本的 ojAlgo？

当我尝试该代码时，我得到一个异常，因为您提供给 tmpSVD.getSolution(...) 方法的“预分配”矩阵是错误的大小/形状。如果您只是删除第二个参数，则分配已为您完成并且代码有效。我得到这个结果：

V[[0.707106781186548, -0.707106781186547], [0.707106781186547, 0.707106781186548]]
Singular values[13.638181696985853, 9.035878689445474E-15]
Solution [0.016129032258064484, 0.01612903225806446]
于 2019-06-25T11:30:15.737 回答