From cc36192f2750d3f727ee3215af4d3654df15b2d6 Mon Sep 17 00:00:00 2001 From: inter Date: Thu, 4 Sep 2025 14:08:50 +0800 Subject: [PATCH] Add File --- .../easyai/matrixTools/MatrixOperation.java | 1020 +++++++++++++++++ 1 file changed, 1020 insertions(+) create mode 100644 src/main/java/org/dromara/easyai/matrixTools/MatrixOperation.java diff --git a/src/main/java/org/dromara/easyai/matrixTools/MatrixOperation.java b/src/main/java/org/dromara/easyai/matrixTools/MatrixOperation.java new file mode 100644 index 0000000..7ab2687 --- /dev/null +++ b/src/main/java/org/dromara/easyai/matrixTools/MatrixOperation.java @@ -0,0 +1,1020 @@ +package org.dromara.easyai.matrixTools; + + +import java.util.*; +import java.util.concurrent.CountDownLatch; +import java.util.concurrent.ExecutorService; +import java.util.concurrent.Executors; + +public class MatrixOperation { + private final int coreNumber; + private ExecutorService POOL;//线程池 + + private static final CudaMatrix cudaMatrix = getCudaMatrix(); + private static final AscendMatrix ascendMatrix = getAscendMatrix(); + + private static AscendMatrix getAscendMatrix() { + ServiceLoader loader = ServiceLoader.load(AscendMatrix.class); + Iterator iterator = loader.iterator(); + if (iterator.hasNext()) { + AscendMatrix cm = iterator.next(); + try { + cm.init(); + return cm; + } catch (Exception e) { + e.printStackTrace(); + } + } + return null; + } + + private static CudaMatrix getCudaMatrix() { + ServiceLoader loader = ServiceLoader.load(CudaMatrix.class); + Iterator iterator = loader.iterator(); + if (iterator.hasNext()) { + CudaMatrix cm = iterator.next(); + try { + cm.init(); + return cm; + } catch (Exception e) { + e.printStackTrace(); + } + } + return null; + } + + public MatrixOperation() { + coreNumber = 1; + } + + public MatrixOperation(int coreNumber) { + this.coreNumber = coreNumber; + if (coreNumber > 1) { + POOL = Executors.newFixedThreadPool(coreNumber); + } + } + + public List rowVectorToList(Matrix matrix) throws Exception { + List list = new ArrayList<>(); + for (int j = 0; j < matrix.getY(); j++) { + list.add(matrix.getNumber(0, j)); + } + return list; + } + + public List addMatrixList(List matrixList1, List matrixList2) throws Exception {//两个矩阵集合相加 + if (matrixList1.size() == matrixList2.size()) { + List result = new ArrayList<>(); + int size = matrixList1.size(); + for (int i = 0; i < size; i++) { + result.add(add(matrixList1.get(i), matrixList2.get(i))); + } + return result; + } else { + throw new Exception("两个矩阵集合相加大小不相等"); + } + } + + //重点 + public Matrix add(Matrix matrix1, Matrix matrix2) throws Exception {//矩阵相加 + if (matrix1.getX() == matrix2.getX() && matrix1.getY() == matrix2.getY()) { + Matrix matrix = new Matrix(matrix1.getX(), matrix1.getY()); + int x = matrix1.getX(); + int y = matrix1.getY(); + for (int i = 0; i < x; i++) {//遍历行 + for (int j = 0; j < y; j++) {//遍历列 + matrix.setNub(i, j, matrix1.getNumber(i, j) + matrix2.getNumber(i, j)); + } + } + return matrix; + } else { + throw new Exception("matrix is not equals"); + } + } + + //重点 + public Matrix addThreeMatrix(Matrix matrix1, Matrix matrix2, Matrix matrix3) throws Exception {//三矩阵相加 + if (matrix1.getX() == matrix2.getX() && matrix1.getY() == matrix2.getY() && matrix1.getX() == matrix3.getX() + && matrix1.getY() == matrix3.getY()) { + Matrix matrix = new Matrix(matrix1.getX(), matrix1.getY()); + int x = matrix1.getX(); + int y = matrix1.getY(); + for (int i = 0; i < x; i++) {//遍历行 + for (int j = 0; j < y; j++) {//遍历列 + matrix.setNub(i, j, matrix1.getNumber(i, j) + matrix2.getNumber(i, j) + matrix3.getNumber(i, j)); + } + } + return matrix; + } else { + throw new Exception("matrix is not equals"); + } + } + + public void center(Matrix matrix) throws Exception {//将矩阵中心化 + float avgValue = matrix.getAVG(); + mathSub(matrix, avgValue); + } + + public float getCrossEntropy(Matrix matrix1, Matrix matrix2) throws Exception {//两个概率矩阵的交叉熵 + float sigMod = 0; + int x = matrix1.getX(); + int y = matrix1.getY(); + if (x == matrix2.getX() && y == matrix2.getY()) { + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + sigMod = sigMod + matrix2.getNumber(i, j) * (float) Math.log(matrix1.getNumber(i, j)); + } + } + return -sigMod; + } else { + throw new Exception("matrix is not equals"); + } + } + + public Matrix softMaxByMatrix(Matrix feature) throws Exception { + float sigma = 0; + int x = feature.getX(); + int y = feature.getY(); + Matrix softMaxMatrix = new Matrix(x, y); + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + float value = feature.getNumber(i, j); + sigma = (float) Math.exp(value) + sigma; + } + } + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + float eSelf = (float) Math.exp(feature.getNumber(i, j)); + float value = (float) (eSelf / sigma); + softMaxMatrix.setNub(i, j, value); + } + } + return softMaxMatrix; + } + + //矩阵相减 重点 + public Matrix sub(Matrix matrix1, Matrix matrix2) throws Exception {// + if (matrix1.getX() == matrix2.getX() && matrix1.getY() == matrix2.getY()) { + Matrix matrix = new Matrix(matrix1.getX(), matrix1.getY()); + int x = matrix1.getX(); + int y = matrix1.getY(); + for (int i = 0; i < x; i++) {//遍历行 + for (int j = 0; j < y; j++) {//遍历列 + matrix.setNub(i, j, matrix1.getNumber(i, j) - matrix2.getNumber(i, j)); + } + } + return matrix; + } else { + throw new Exception("matrix is not equals"); + } + } + + private int getLBPValue(Matrix matrix) throws Exception { + int value = 0; + float avg = matrix.getAVG(); + for (int i = 0; i < 3; i++) { + for (int j = 0; j < 3; j++) { + if (i != 1 || j != 1) { + value = value << 1; + if (matrix.getNumber(i, j) > avg) { + value = value | 1; + } + } + } + } + return value; + } + + public Matrix lbpMatrix(Matrix addMatrix) throws Exception { + int x = addMatrix.getX(); + int y = addMatrix.getY(); + Matrix matrix = new Matrix(x / 3, y / 3); + for (int i = 0; i <= x - 3; i += 3) { + for (int j = 0; j <= y - 3; j += 3) { + Matrix aMatrix = addMatrix.getSonOfMatrix(i, j, 3, 3); + int lbp = getLBPValue(aMatrix); + matrix.setNub(i / 3, j / 3, lbp); + } + } + return matrix; + } + + //多元线性回归 + public Matrix getLinearRegression(Matrix parameter, Matrix out) throws Exception { + if (parameter.getX() == out.getX() && out.isVector()) { + //将参数矩阵转置 + Matrix matrix1 = transPosition(parameter); + //转置的参数矩阵乘以参数矩阵 + Matrix matrix2 = mulMatrix(matrix1, parameter); + //求上一步的逆矩阵 这一步需要矩阵非奇异,若出现奇异矩阵，则返回0矩阵，意味失败 + Matrix matrix3 = getInverseMatrix(matrix2); + if (matrix3.getX() == 1 && matrix3.getY() == 1) { + return matrix3; + } else { + //逆矩阵乘以转置矩阵 + Matrix matrix4 = mulMatrix(matrix3, matrix1); + //最后乘以输出矩阵,生成权重矩阵并返回 + return mulMatrix(matrix4, out); + } + } else { + throw new Exception("invalid regression matrix"); + } + } + + public float getEDistByMatrix(Matrix matrix1, Matrix matrix2) throws Exception { + if (matrix1.getX() == matrix2.getX() && matrix1.getY() == matrix2.getY()) { + int x = matrix1.getX(); + int y = matrix1.getY(); + float sigma = 0; + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + float sub = matrix1.getNumber(i, j) - matrix2.getNumber(i, j); + sigma = sigma + (float) Math.pow(sub, 2); + } + } + return sigma / (x * y); + } else { + throw new Exception("two matrixes is not equals"); + } + } + + //返回两个向量之间的欧氏距离的平方 + public float getEDist(Matrix matrix1, Matrix matrix2) throws Exception { + if (matrix1.isRowVector() && matrix2.isRowVector() && matrix1.getY() == matrix2.getY()) { + Matrix matrix = sub(matrix1, matrix2); + return getNorm(matrix); + } else { + throw new Exception("this matrix is not rowVector or length different"); + } + } + + public float errorNub(Matrix matrix, Matrix avgMatrix) throws Exception {//求均方误差 + int y = matrix.getY(); + if (matrix.isRowVector() && avgMatrix.isRowVector() && y == avgMatrix.getY()) { + float[] subAll = new float[y]; + for (int j = 0; j < y; j++) { + float mySelf = matrix.getNumber(0, j); + float avg = avgMatrix.getNumber(0, j); + float sub = (float) Math.pow(avg - mySelf, 2); + subAll[j] = sub; + } + float sigma = 0; + for (int i = 0; i < y; i++) { + sigma = sigma + subAll[i]; + } + return sigma / y; + } else { + throw new Exception("this matrix is not rowVector or length different"); + } + } + + public Matrix pushVector(Matrix myMatrix, Matrix matrix, boolean addRow) throws Exception { + //向一个矩阵里合并一个行向量或者列向量到矩阵行或者列的末尾 + if (matrix.getX() == 1 || matrix.getY() == 1) { + Matrix addMatrix; + if (addRow) {//增加一行 + if (matrix.getY() != myMatrix.getY()) { + throw new Exception("this matrix column is not equals"); + } + addMatrix = new Matrix(myMatrix.getX() + 1, myMatrix.getY()); + } else {//增加一列 + if (matrix.getX() != myMatrix.getX()) { + throw new Exception("this matrix row is not equals"); + } + addMatrix = new Matrix(myMatrix.getX(), myMatrix.getY() + 1); + } + for (int i = 0; i < addMatrix.getX(); i++) { + for (int j = 0; j < addMatrix.getY(); j++) { + if (addRow) { + if (i == addMatrix.getX() - 1) {//最后一行 + addMatrix.setNub(i, j, matrix.getNumber(0, j)); + } else { + addMatrix.setNub(i, j, myMatrix.getNumber(i, j)); + } + } else { + if (j == addMatrix.getY() - 1) {//最后一列 + addMatrix.setNub(i, j, matrix.getNumber(i, 0)); + } else { + addMatrix.setNub(i, j, myMatrix.getNumber(i, j)); + } + } + } + } + return addMatrix; + } else { + throw new Exception("this matrix is not a vector"); + } + } + + public Matrix push(Matrix matrix, float nub, boolean isRow) throws Exception {//向一个向量里PUSH一个值 + if (matrix.getX() == 1 || matrix.getY() == 1) { + Matrix myMatrix; + int nubs; + if (isRow) {//行向量 + nubs = matrix.getY() + 1; + myMatrix = new Matrix(1, nubs); + } else {//列向量 + nubs = matrix.getX() + 1; + myMatrix = new Matrix(nubs, 1); + } + for (int i = 0; i < nubs; i++) { + if (i == nubs - 1) { + if (isRow) { + myMatrix.setNub(0, i, nub); + } else { + myMatrix.setNub(i, 0, nub); + } + } else { + if (isRow) {//行向量 + myMatrix.setNub(0, i, matrix.getNumber(0, i)); + } else {//列向量 + myMatrix.setNub(i, 0, matrix.getNumber(i, 0)); + } + } + } + return myMatrix; + } else { + throw new Exception("this matrix is not a vector"); + } + } + + public Matrix getPoolVector(Matrix matrix) throws Exception { + if (matrix.getX() == 1 || matrix.getY() == 1) { + Matrix vector; + int nub; + boolean isRow = false; + if (matrix.getX() == 1) {//行向量 + isRow = true; + nub = matrix.getY() / 4; + vector = new Matrix(1, nub); + } else {//列向量 + nub = matrix.getX() / 4; + vector = new Matrix(nub, 1); + } + int k = 0; + for (int i = 0; i < nub * 4 - 3; i += 4) { + float max = 0; + if (isRow) { + max = matrix.getNumber(0, i); + max = getMax(max, matrix.getNumber(0, i + 1)); + max = getMax(max, matrix.getNumber(0, i + 2)); + max = getMax(max, matrix.getNumber(0, i + 3)); + vector.setNub(0, k, max); + } else { + max = matrix.getNumber(i, 0); + max = getMax(max, matrix.getNumber(i + 1, 0)); + max = getMax(max, matrix.getNumber(i + 2, 0)); + max = getMax(max, matrix.getNumber(i + 3, 0)); + vector.setNub(k, 0, max); + } + k++; + } + return vector; + } else { + throw new Exception("this matrix is not a vector"); + } + + } + + private float getMax(float o1, float o2) { + return (float) Math.max(o1, o2); + } + + public Matrix matrixToVector(Matrix matrix, boolean isRow) throws Exception {//将一个矩阵转成向量 + int x = matrix.getX(); + int y = matrix.getY(); + Matrix myMatrix; + if (isRow) { + myMatrix = new Matrix(1, x * y); + } else { + myMatrix = new Matrix(x * y, 1); + } + int t = 0; + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + if (isRow) { + myMatrix.setNub(0, t, matrix.getNumber(i, j)); + } else { + myMatrix.setNub(t, 0, matrix.getNumber(i, j)); + } + t++; + } + } + return myMatrix; + } + + public float innerProduct(Matrix matrix1, Matrix matrix2) throws Exception {//两个向量内积 + if (matrix1.getX() == matrix2.getX() && matrix1.getY() == matrix2.getY()) { + float sigma = 0; + for (int i = 0; i < matrix1.getX(); i++) { + for (int j = 0; j < matrix1.getY(); j++) { + sigma = sigma + matrix1.getNumber(i, j) * matrix2.getNumber(i, j); + } + } + return sigma; + } else { + throw new Exception("两个向量的长宽必须相同"); + } + } + + public float getNorm(Matrix matrix) throws Exception {//求向量范数 + if (matrix.getY() == 1 || matrix.getX() == 1) { + float nub = 0; + for (int i = 0; i < matrix.getX(); i++) { + for (int j = 0; j < matrix.getY(); j++) { + nub = (float) Math.pow(matrix.getNumber(i, j), 2) + nub; + } + } + return (float) Math.sqrt(nub); + } else { + throw new Exception("this matrix is not vector"); + } + } + + public float getNormCos(Matrix matrix1, Matrix matrix2) throws Exception {//求两个向量之间的余弦 + float inner = innerProduct(matrix1, matrix2); + float mulNorm = getNorm(matrix1) * getNorm(matrix2); + return inner / mulNorm; + } + + public Matrix transPosition(Matrix matrix) throws Exception {//矩阵转置 + Matrix myMatrix = new Matrix(matrix.getY(), matrix.getX()); + for (int i = 0; i < matrix.getY(); i++) { + Matrix matrixColumn = matrix.getColumn(i); + for (int j = 0; j < matrixColumn.getX(); j++) { + float myNode = matrixColumn.getNumber(j, 0); + myMatrix.setNub(i, j, myNode); + } + } + return myMatrix; + } + + public float convolution(Matrix matrix, Matrix kernel, int x, int y) throws Exception {//计算卷积 + float allNub = 0; + int xr = 0; + int yr = 0; + int kxMax = kernel.getX(); + int kyMax = kernel.getY(); + for (int i = 0; i < kxMax; i++) { + xr = i + x; + for (int j = 0; j < kyMax; j++) { + yr = j + y; + allNub = matrix.getNumber(xr, yr) * kernel.getNumber(i, j) + allNub; + } + } + + return allNub; + } + + public float getKernelNub(Matrix matrix, Matrix kernel) throws Exception { + float allNub = 0; + int x = matrix.getX(); + int y = matrix.getY(); + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + allNub = matrix.getNumber(i, j) * kernel.getNumber(i, j) + allNub; + } + } + return allNub; + } + + public int inverseNumber(float[] myInverse) {//逆序数奇偶性判定 + int size = myInverse.length; + int inverserNumber = 0; + for (int i = 0; i < size; i++) { + float element = myInverse[i]; + for (int j = (i + 1); j < size; j++) { + if (myInverse[j] < element) { + inverserNumber++; + } + } + } + return inverserNumber; + } + + public Matrix getInverseMatrix(Matrix matrix) throws Exception {//矩阵求逆 + float def = matrix.getDet(); + if (def != 0) { + def = 1 / def; + Matrix myMatrix = adjointMatrix(matrix);//伴随矩阵 + mathMul(myMatrix, def); + return myMatrix; + } else { + //System.out.println("matrix def is zero error:"); + //System.out.println(matrix.getString()); + return new Matrix(1, 1); + } + } + + public Matrix adjointMatrix(Matrix matrix) throws Exception {//求伴随矩阵 + Matrix myMatrix = new Matrix(matrix.getX(), matrix.getY()); + for (int i = 0; i < matrix.getX(); i++) { + for (int j = 0; j < matrix.getY(); j++) { + myMatrix.setNub(i, j, algebraicCofactor(matrix, i, j)); + } + } + + return transPosition(myMatrix); + } + + public float algebraicCofactor(Matrix matrix, int row, int column) throws Exception {//获取代数余子式 + if (row >= 0 && column >= 0 && row < matrix.getX() && column < matrix.getY()) { + int x = matrix.getX() - 1; + int y = matrix.getY() - 1; + int ij = row + column + 2; + int oldX = 0;//从目标矩阵中取值的X + int oldY = 0;//从目标矩阵中取值的Y + boolean isXNext = false;//行是否跳过一次 + Matrix myMatrix = new Matrix(x, y); + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + if (i == row && !isXNext) { + isXNext = true; + oldX++; + } + if (j == column) { + oldY++; + } + myMatrix.setNub(i, j, matrix.getNumber(oldX, oldY)); + oldY++; + } + oldX++; + oldY = 0; + } + float dm = myMatrix.getDet(); + if ((ij % 2) != 0) {//ij是奇数 + dm = -dm; + } + return dm; + } else { + throw new Exception("row or column index Beyond the limit"); + } + } + + public Matrix matrixSoftMaxPd(Matrix qkt, Matrix errorMatrix, float wordVectorDimension) throws Exception {//重重点 + if (null != cudaMatrix) { + return cudaMatrix.matrixSoftMaxPd(qkt, errorMatrix, wordVectorDimension); + } + + float param = (float) Math.sqrt(wordVectorDimension); + int x = qkt.getX(); + int y = qkt.getY(); + Matrix grMatrix = new Matrix(x, y); + for (int i = 0; i < x; i++) { + Matrix qr = qkt.getRow(i);//该行的行向量 + for (int j = 0; j < y; j++) { + float jValue = qr.getNumber(0, j);//遍历qr每一个元素，分别对他们求偏导 + int z = qr.getY(); + float sigma = 0; + for (int k = 0; k < z; k++) { + float kValue = qr.getNumber(0, k);//遍历qr每一个元素，分别对他们求偏导 + float error = errorMatrix.getNumber(i, k); + float er; + if (k != j) { + er = -error * kValue * jValue; + } else { + er = jValue * (1 - jValue) * error; + } + sigma = sigma + er; + } + float gr = (float) (sigma / param);//该位置梯度 + grMatrix.setNub(i, j, gr); + } + } + return grMatrix; + } + + public void softMax(Matrix matrix) throws Exception {//重重点 + for (int i = 0; i < matrix.getX(); i++) { + Matrix row = matrix.getRow(i); + float sigma = getRowSoftMaxSigma(row); + for (int j = 0; j < matrix.getY(); j++) { + float self = row.getNumber(0, j); + float eSelf = (float) Math.exp(self); + matrix.setNub(i, j, eSelf / sigma); + } + } + } + + private float getRowSoftMaxSigma(Matrix row) throws Exception { + float sigma = 0; + for (int i = 0; i < row.getY(); i++) { + float value = row.getNumber(0, i); + sigma = (float) Math.exp(value) + sigma; + } + return sigma; + } + + public Matrix matrixPointDiv(Matrix matrix1, Matrix matrix2) throws Exception {//矩阵点除 + int x = matrix1.getX(); + int y = matrix1.getY(); + Matrix matrix = new Matrix(x, y); + if (matrix2.getX() == x && matrix2.getY() == y) { + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + matrix.setNub(i, j, matrix1.getNumber(i, j) / matrix2.getNumber(i, j)); + } + } + } else { + throw new Exception("two matrix is not equals"); + } + return matrix; + } + + public Matrix matrixMulPd(Matrix errorMatrix, Matrix first, Matrix second, boolean isFirstPd) throws Exception {//对两个相乘的矩阵求偏导 + Matrix matrix; + if (isFirstPd) {//对相乘的前矩阵进行求导 + matrix = mulMatrix(errorMatrix, transPosition(second)); + } else { + matrix = mulMatrix(transPosition(first), errorMatrix); + } + return matrix; + } + + //重点 + public float getSdByMatrix(Matrix m, float avg, float e) throws Exception {//计算矩阵元素的标准差 + float var = 0; + float size = m.getX() * m.getY(); + int x = m.getX(); + int y = m.getY(); + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + var = var + (float) Math.pow(m.getNumber(i, j) - avg, 2); + } + } + return (float) Math.sqrt(var / size + e); + } + + private Matrix mulMatrixOne(Matrix matrix1, Matrix matrix2) throws Exception {//矩阵相乘单进程 + if (matrix1.getY() == matrix2.getX()) { + Matrix matrix = new Matrix(matrix1.getX(), matrix2.getY()); + for (int i = 0; i < matrix1.getX(); i++) { + Matrix matrixRow = matrix1.getRow(i);//行向量 + for (int j = 0; j < matrix2.getY(); j++) { + Matrix matrixColumn = matrix2.getColumn(j); + float columnAllNumber = 0;//对每一项的乘积求和 + for (int h = 0; h < matrixColumn.getX(); h++) { + float columnNumber = matrixColumn.getNumber(h, 0); + float rowNumber = matrixRow.getNumber(0, h); + float nowNumber = columnNumber * rowNumber; + columnAllNumber = columnAllNumber + nowNumber; + } + if (!Float.isNaN(columnAllNumber)) { + matrix.setNub(i, j, columnAllNumber); + } else { + System.out.println(matrix1); + System.out.println(matrix2); + throw new Exception("计算时出现数值溢出，梯度爆炸!"); + } + } + } + return matrix; + } else { + throw new Exception("row is not equals column"); + } + } + + private Matrix mulMatrixMany(Matrix matrix1, Matrix matrix2) throws Exception {//矩阵相乘多进程 + if (matrix1.getY() == matrix2.getX()) { + Matrix matrix = new Matrix(matrix1.getX(), matrix2.getY()); + CountDownLatch countDownLatch = new CountDownLatch(matrix1.getX() * matrix2.getY()); + int x = matrix1.getX(); + int y = matrix2.getY(); + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + MatrixMulAccelerate matrixMulAccelerate = new MatrixMulAccelerate(matrix1, matrix2 + , matrix, i, j, countDownLatch); + POOL.execute(matrixMulAccelerate); + } + } + countDownLatch.await(); + return matrix; + } else { + throw new Exception("row is not equals column"); + } + } + + public List mulMatrixList(List matrixList1, List matrixList2) throws Exception {//两个矩阵集合相乘 + if (matrixList1.size() == matrixList2.size()) { + List mulMatrixList = new ArrayList<>(); + int size = matrixList1.size(); + for (int i = 0; i < size; i++) { + Matrix mulMatrix = mulMatrix(matrixList1.get(i), matrixList2.get(i)); + mulMatrixList.add(mulMatrix); + } + return mulMatrixList; + } else { + throw new IllegalArgumentException("两个矩阵集合的大小不相等"); + } + } + + //重重点 + public Matrix mulMatrix(Matrix matrix1, Matrix matrix2) throws Exception {//矩阵相乘 + if (null != cudaMatrix) { + return cudaMatrix.mulMatrix(matrix1, matrix2); + } else if (null != ascendMatrix) { + return ascendMatrix.mulMatrix(matrix1, matrix2); + } + + if (coreNumber > 1) { + return mulMatrixMany(matrix1, matrix2); + } else { + return mulMatrixOne(matrix1, matrix2); + } + } + + //重重点 + public Matrix mathMulBySelf(Matrix matrix, float nub) throws Exception {//矩阵数乘并返回新矩阵 + int x = matrix.getX(); + int y = matrix.getY(); + Matrix myMatrix = new Matrix(x, y); + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + myMatrix.setNub(i, j, matrix.getNumber(i, j) * nub); + } + } + return myMatrix; + } + + public void mathMulByList(List matrixList, float value) throws Exception { + for (Matrix matrix : matrixList) { + mathMul(matrix, value); + } + } + + //重点 + public void mathMul(Matrix matrix, float nub) throws Exception {//矩阵数乘 + for (int i = 0; i < matrix.getX(); i++) { + for (int j = 0; j < matrix.getY(); j++) { + matrix.setNub(i, j, matrix.getNumber(i, j) * nub); + } + } + } + + //重点 + public void mathAdd(Matrix matrix, float nub) throws Exception {//矩阵数加 + for (int i = 0; i < matrix.getX(); i++) { + for (int j = 0; j < matrix.getY(); j++) { + matrix.setNub(i, j, matrix.getNumber(i, j) + nub); + } + } + } + + //重点 + public void mathSub(Matrix matrix, float nub) throws Exception {//矩阵数减 + for (int i = 0; i < matrix.getX(); i++) { + for (int j = 0; j < matrix.getY(); j++) { + matrix.setNub(i, j, matrix.getNumber(i, j) - nub); + } + } + } + + //重点 + public void mathDiv(Matrix matrix, float nub) throws Exception {//矩阵数除 + for (int i = 0; i < matrix.getX(); i++) { + for (int j = 0; j < matrix.getY(); j++) { + matrix.setNub(i, j, matrix.getNumber(i, j) / nub); + } + } + } + + //矩阵转LIST + public List matrixToList(Matrix matrix) throws Exception { + List list = new ArrayList<>(); + int x = matrix.getX(); + int y = matrix.getY(); + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + list.add(matrix.getNumber(i, j)); + } + } + return list; + } + + public Matrix vectorToMatrix(Matrix vector, int xSize, int ySize) throws Exception {//向量转矩阵 + Matrix matrix = new Matrix(xSize, ySize); + if (vector.isRowVector()) {//行向量 + for (int i = 0; i < xSize; i++) { + for (int j = 0; j < ySize; j++) { + int index = i * ySize + j; + matrix.setNub(i, j, vector.getNumber(0, index)); + } + } + } else {//列向量 + for (int i = 0; i < xSize; i++) { + for (int j = 0; j < ySize; j++) { + int index = i * ySize + j; + matrix.setNub(i, j, vector.getNumber(index, 0)); + } + } + } + return matrix; + } + + //list转矩阵 + public Matrix ListToMatrix(List list, int matrixX, int matrixY) throws Exception { + Matrix matrix = new Matrix(matrixX, matrixY); + for (int i = 0; i < matrixX; i++) { + for (int j = 0; j < matrixY; j++) { + int index = i * matrixY + j; + matrix.setNub(i, j, list.get(index)); + } + } + return matrix; + } + + //list转行向量 + public Matrix listToRowVector(List list) throws Exception { + Matrix matrix = new Matrix(1, list.size()); + for (int i = 0; i < list.size(); i++) { + matrix.setNub(0, i, list.get(i)); + } + return matrix; + } + + //list转行向量定长 + public Matrix listToRowVector(List list, int nub) throws Exception { + Matrix matrix = new Matrix(1, nub); + for (int i = 0; i < nub; i++) { + float n = 0; + if (list.size() > i) { + n = list.get(i); + } + matrix.setNub(0, i, n); + } + return matrix; + } + + private void inputVector(Matrix matrix, Matrix feature, int index, int kenLen) throws Exception { + int y = matrix.getY(); + for (int i = 0; i < y; i++) { + matrix.setNub(index, i, feature.getNumber(i / kenLen, i % kenLen)); + } + } + + //重重点 + public Matrix im2col(Matrix matrix, int kernLen, int step) throws Exception {//卷积拉平 + int ySize = kernLen * kernLen; + int x = matrix.getX(); + int y = matrix.getY(); + int xSize = ((x - (kernLen - step)) / step) * ((y - (kernLen - step)) / step); + Matrix myMatrix = new Matrix(xSize, ySize); + int index = 0; + for (int i = 0; i <= x - kernLen; i += step) { + for (int j = 0; j <= y - kernLen; j += step) { + inputVector(myMatrix, matrix.getSonOfMatrix(i, j, kernLen, kernLen), index, kernLen); + index++; + } + } + return myMatrix; + } + + //重重点 + public Matrix reverseIm2col(Matrix matrix, int kernLen, int step, int xSize, int ySize) throws Exception {//逆向im2col + int col = matrix.getY();//核长 + int row = matrix.getX(); + int sub = kernLen - step; + int y = (ySize - sub) / step;//线性变换后矩阵的列数,一行该有几列 + Matrix myMatrix = new Matrix(xSize, ySize); + for (int i = 0; i < row; i++) {//某一行全部的值,首先要看这一行的值应该在逆运算的第几行 + int xz = (i / y) * step;//左上角行数 + int yz = (i % y) * step;//左上角列数 + for (int j = 0; j < col; j++) { + int xr = j / kernLen + xz; + int yr = j % kernLen + yz; + float value = myMatrix.getNumber(xr, yr) + matrix.getNumber(i, j); + myMatrix.setNub(xr, yr, value); + } + } + return myMatrix; + } + + private void insertVectorValue(Matrix matrix, Matrix vector, int col) throws Exception {//向一个矩阵指定行注入一个向量 + int x = matrix.getX(); + int y = matrix.getY(); + if (x == vector.getX() && vector.isVector() && y > col) {//矩阵为行向量 + for (int i = 0; i < x; i++) { + matrix.setNub(i, col, vector.getNumber(i, 0)); + } + } else { + throw new Exception("注入向量异常，注入的必须为向量，且矩阵的列数必须与向量的列数相等，且行数没有溢出"); + } + } + + private void matrixSqrt(Matrix matrix) throws Exception { + int x = matrix.getX(); + int y = matrix.getY(); + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + if (i == j) { + float value = (float) Math.sqrt(matrix.getNumber(i, j)); + matrix.setNub(i, j, (float) value); + } + } + } + } + + public Matrix pca(Matrix feature, Matrix avgMatrix, int time, int featureSize) throws Exception { + if (feature.getX() == avgMatrix.getX() && feature.getY() == avgMatrix.getY()) { + Matrix myMatrix = sub(feature, avgMatrix); + return svd(myMatrix, time, featureSize).getMatrixU(); + } else { + throw new Exception("特征矩阵与平均值矩阵大小不相等"); + } + } + + public SVDBody svd(Matrix matrix, int time, int featureSize) throws Exception {//svd分解 + Matrix tMatrix = transPosition(matrix);//转制矩阵 + Matrix UA = mulMatrix(matrix, tMatrix); + Matrix VA = mulMatrix(tMatrix, matrix); + FeatureDe featureU = qrIteration(UA, time); + FeatureDe featureV = qrIteration(VA, time); + Matrix uValue = featureU.getFeatureValue();//特征值 + Matrix U = featureU.getFeatureMatrix(); + Matrix VT = transPosition(featureV.getFeatureMatrix()); + if (featureSize <= U.getY() && featureSize <= VT.getX()) { + SVDBody svdBody = new SVDBody(); + Matrix rU = U.getSonOfMatrix(0, 0, U.getX(), featureSize); + Matrix rVT = VT.getSonOfMatrix(0, 0, featureSize, VT.getY()); + Matrix feature = uValue.getSonOfMatrix(0, 0, featureSize, featureSize); + matrixSqrt(feature); + svdBody.setMatrixU(rU); + svdBody.setMatrixVT(rVT); + svdBody.setFeature(feature); + return svdBody; + } else { + throw new Exception("指定尺寸过大"); + } + } + + public FeatureDe qrIteration(Matrix matrix, int time) throws Exception {//QR迭代求特征值 + Matrix V = null; + Matrix featureValue = matrix.copy(); + for (int i = 0; i < time; i++) { + QRMatrix qrMatrix = qrd(featureValue); + Matrix matrixQ = qrMatrix.getQ(); + Matrix matrixR = qrMatrix.getR(); + if (V == null) { + V = matrixQ; + } else { + V = mulMatrix(V, matrixQ); + } + featureValue = mulMatrix(matrixR, matrixQ); + } + int x = featureValue.getX(); + int y = featureValue.getY(); + Matrix featureMatrix = new Matrix(x, y); + for (int i = 0; i < x; i++) { + for (int j = 0; j < y; j++) { + if (i == j) { + featureMatrix.setNub(i, j, featureValue.getNumber(i, j)); + } + } + } + FeatureDe featureDe = new FeatureDe(); + featureDe.setFeatureValue(featureMatrix); + featureDe.setFeatureMatrix(V); + return featureDe; + } + + public QRMatrix qrd(Matrix matrix) throws Exception {//返回qr分解 + int x = matrix.getX(); + int y = matrix.getY(); + if (x == y) { + Matrix schMatrix = new Matrix(x, y); + Matrix R = new Matrix(x, x);//R矩阵 + Matrix normMatrix = new Matrix(x, x); + for (int i = 0; i < y; i++) { + Matrix xn = matrix.getColumn(i); + R.setNub(i, i, 1F); + if (i > 0) { + for (int k = 0; k < i; k++) { + Matrix vn = schMatrix.getColumn(k); + float value = innerProduct(xn, vn) / innerProduct(vn, vn); + R.setNub(k, i, value); + mathMul(vn, value); + xn = sub(xn, vn); + } + } + float norm = getNorm(xn);//范数 + normMatrix.setNub(i, i, norm); + insertVectorValue(schMatrix, xn, i); + } + R = mulMatrix(normMatrix, R); + for (int i = 0; i < x; i++) { + float norm = normMatrix.getNumber(i, i); + for (int j = 0; j < y; j++) { + float value = schMatrix.getNumber(j, i) / norm; + schMatrix.setNub(j, i, value); + } + } + QRMatrix qrMatrix = new QRMatrix(); + qrMatrix.setR(R); + qrMatrix.setQ(schMatrix); + return qrMatrix; + } else { + throw new Exception("请将矩阵先处理为方阵才可以进行QR分解"); + } + } + +}