org.nd4j.linalg.factory.Nd4j.getBlasWrapper()方法的使用及代码示例

x33g5p2x  于2022-01-24 转载在 其他  
字(10.3k)|赞(0)|评价(0)|浏览(163)

本文整理了Java中org.nd4j.linalg.factory.Nd4j.getBlasWrapper()方法的一些代码示例，展示了Nd4j.getBlasWrapper()的具体用法。这些代码示例主要来源于Github/Stackoverflow/Maven等平台，是从一些精选项目中提取出来的代码，具有较强的参考意义，能在一定程度帮忙到你。Nd4j.getBlasWrapper()方法的具体详情如下：
包路径：org.nd4j.linalg.factory.Nd4j
类名称：Nd4j
方法名：getBlasWrapper

Nd4j.getBlasWrapper介绍

暂无

代码示例

代码示例来源：origin: deeplearning4j/dl4j-examples

  1. public static void main(String[] args) {
  3.     Nd4j.setDataType(DataBuffer.Type.DOUBLE);
  4.     INDArray arr = Nd4j.create(300);
  5.     double numTimes = 10000000;
  6.     double total = 0;
  8.     for(int i = 0; i < numTimes; i++) {
  9.       long start = System.nanoTime();
  10.       Nd4j.getBlasWrapper().axpy(new Integer(1), arr,arr);
  11.       long after = System.nanoTime();
  12.       long add = Math.abs(after - start);
  13.       System.out.println("Took " + add);
  14.       total += Math.abs(after - start);
  15.     }
  16.     System.out.println("Avg time " + (total / numTimes));
  17.   }
  18. }

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2. public Map<Integer, Double> labelCounts() {
  3.   Map<Integer, Double> ret = new HashMap<>();
  4.   if (labels == null)
  5.     return ret;
  6.   long nTensors = labels.tensorssAlongDimension(1);
  7.   for (int i = 0; i < nTensors; i++) {
  8.     INDArray row = labels.tensorAlongDimension(i, 1);
  9.     INDArray javaRow = labels.javaTensorAlongDimension(i, 1);
  10.     int maxIdx = Nd4j.getBlasWrapper().iamax(row);
  11.     int maxIdxJava = Nd4j.getBlasWrapper().iamax(javaRow);
  12.     if (maxIdx < 0)
  13.       throw new IllegalStateException("Please check the iamax implementation for "
  14.               + Nd4j.getBlasWrapper().getClass().getName());
  15.     if (ret.get(maxIdx) == null)
  16.       ret.put(maxIdx, 1.0);
  17.     else
  18.       ret.put(maxIdx, ret.get(maxIdx) + 1.0);
  19.   }
  20.   return ret;
  21. }

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2. public int outcome() {
  3.   return Nd4j.getBlasWrapper().iamax(getLabels());
  4. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Scale by 1 / norm2 of the matrix
  3.  *
  4.  * @param toScale the ndarray to scale
  5.  * @return the scaled ndarray
  6.  */
  7. public static INDArray unitVec(INDArray toScale) {
  8.   double length = toScale.norm2Number().doubleValue();
  9.   if (length > 0) {
  10.     if (toScale.data().dataType() == (DataBuffer.Type.FLOAT))
  11.       return Nd4j.getBlasWrapper().scal(1.0f / (float) length, toScale);
  12.     else
  13.       return Nd4j.getBlasWrapper().scal(1.0 / length, toScale);
  14.   }
  15.   return toScale;
  16. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Compute generalized eigenvalues of the problem A x = L x.
  3.  * Matrix A is modified in the process, holding eigenvectors after execution.
  4.  *
  5.  * @param A symmetric Matrix A. After execution, A will contain the eigenvectors as columns
  6.  * @return a vector of eigenvalues L.
  7.  */
  8. public static INDArray symmetricGeneralizedEigenvalues(INDArray A) {
  9.   INDArray eigenvalues = Nd4j.create(A.rows());
  10.   Nd4j.getBlasWrapper().syev('V', 'L', A, eigenvalues);
  11.   return eigenvalues;
  12. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /** Matrix multiply: Implements c = alpha*op(a)*op(b) + beta*c where op(X) means transpose X (or not)
  2.  * depending on setting of arguments transposeA and transposeB.<br>
  3.  * Note that matrix c MUST be fortran order, have zero offset and have c.data().length == c.length().
  4.  * An exception will be thrown otherwise.<br>
  5.  * Don't use this unless you know about level 3 blas and NDArray storage orders.
  6.  * @param a First matrix
  7.  * @param b Second matrix
  8.  * @param c result matrix. Used in calculation (assuming beta != 0) and result is stored in this. f order,
  9.  *          zero offset and length == data.length only
  10.  * @param transposeA if true: transpose matrix a before mmul
  11.  * @param transposeB if true: transpose matrix b before mmul
  12.  * @return result, i.e., matrix c is returned for convenience
  13.  */
  14. public static INDArray gemm(INDArray a,
  15.               INDArray b,
  16.               INDArray c,
  17.               boolean transposeA,
  18.               boolean transposeB,
  19.               double alpha,
  20.               double beta) {
  21.   getBlasWrapper().level3().gemm(a, b, c, transposeA, transposeB, alpha, beta);
  22.   return c;
  23. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * in place subtraction of two matrices
  3.  *
  4.  * @param other  the second ndarray to subtract
  5.  * @param result the result ndarray
  6.  * @return the result of the subtraction
  7.  */
  8. @Override
  9. public IComplexNDArray subi(INDArray other, INDArray result) {
  10.   IComplexNDArray cOther = (IComplexNDArray) other;
  11.   IComplexNDArray cResult = (IComplexNDArray) result;
  12.   if (other.isScalar())
  13.     return subi(cOther.getComplex(0), result);
  14.   if (result == this)
  15.     Nd4j.getBlasWrapper().axpy(Nd4j.NEG_UNIT, cOther, cResult);
  16.   else if (result == other) {
  17.     if (data.dataType() == (DataBuffer.Type.DOUBLE)) {
  18.       Nd4j.getBlasWrapper().scal(Nd4j.NEG_UNIT.asDouble(), cResult);
  19.       Nd4j.getBlasWrapper().axpy(Nd4j.UNIT, this, cResult);
  20.     } else {
  21.       Nd4j.getBlasWrapper().scal(Nd4j.NEG_UNIT.asFloat(), cResult);
  22.       Nd4j.getBlasWrapper().axpy(Nd4j.UNIT, this, cResult);
  23.     }
  24.   } else {
  25.     Nd4j.getBlasWrapper().copy(this, result);
  26.     Nd4j.getBlasWrapper().axpy(Nd4j.NEG_UNIT, cOther, cResult);
  27.   }
  28.   return cResult;
  29. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Compute generalized eigenvalues of the problem A x = L x.
  3.  * Matrix A is modified in the process, holding eigenvectors as columns after execution.
  4.  *
  5.  * @param A symmetric Matrix A. After execution, A will contain the eigenvectors as columns
  6.  * @param calculateVectors if false, it will not modify A and calculate eigenvectors
  7.  * @return a vector of eigenvalues L.
  8.  */
  9. public static INDArray symmetricGeneralizedEigenvalues(INDArray A, boolean calculateVectors) {
  10.   INDArray eigenvalues = Nd4j.create(A.rows());
  11.   Nd4j.getBlasWrapper().syev('V', 'L', (calculateVectors ? A : A.dup()), eigenvalues);
  12.   return eigenvalues;
  13. }

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2. public INDArray mmul(INDArray other) {
  3.   long[] shape = {rows(), other.columns()};
  4.   INDArray result = createUninitialized(shape, 'f');
  5.   if (result.isScalar())
  6.     return Nd4j.scalar(Nd4j.getBlasWrapper().dot(this, other));
  7.   return mmuli(other, result);
  8. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Computes the eigenvalues of a general matrix.
  3.  */
  4. public static IComplexNDArray eigenvalues(INDArray A) {
  5.   assert A.rows() == A.columns();
  6.   INDArray WR = Nd4j.create(A.rows(), A.rows());
  7.   INDArray WI = WR.dup();
  8.   Nd4j.getBlasWrapper().geev('N', 'N', A.dup(), WR, WI, dummy, dummy);
  9.   return Nd4j.createComplex(WR, WI);
  10. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Perform a copy matrix multiplication
  3.  *
  4.  * @param other the other matrix to perform matrix multiply with
  5.  * @return the result of the matrix multiplication
  6.  */
  7. @Override
  8. public INDArray mmul(INDArray other) {
  9.   // FIXME: for 1D case, we probably want vector output here?
  10.   long[] shape = {rows(), other.rank() == 1 ? 1 : other.columns()};
  11.   INDArray result = createUninitialized(shape, 'f');
  12.   if (result.isScalar())
  13.     return Nd4j.scalar(Nd4j.getBlasWrapper().dot(this, other));
  14.   return mmuli(other, result);
  15. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Compute generalized eigenvalues of the problem A x = L B x.
  3.  * The data will be unchanged, no eigenvectors returned.
  4.  *
  5.  * @param A symmetric Matrix A.
  6.  * @param B symmetric Matrix B.
  7.  * @return a vector of eigenvalues L.
  8.  */
  9. public static INDArray symmetricGeneralizedEigenvalues(INDArray A, INDArray B) {
  10.   assert A.rows() == A.columns();
  11.   assert B.rows() == B.columns();
  12.   INDArray W = Nd4j.create(A.rows());
  13.   A = InvertMatrix.invert(B, false).mmuli(A);
  14.   Nd4j.getBlasWrapper().syev('V', 'L', A, W);
  15.   return W;
  16. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Returns a column vector where each entry is the nth bilinear
  3.  * product of the nth slices of the two tensors.
  4.  */
  5. @Override
  6. public INDArray bilinearProducts(INDArray curr, INDArray in) {
  7.   assert curr.shape().length == 3;
  8.   if (in.columns() != 1) {
  9.     throw new AssertionError("Expected a column vector");
  10.   }
  11.   if (in.rows() != curr.size(curr.shape().length - 1)) {
  12.     throw new AssertionError("Number of rows in the input does not match number of columns in tensor");
  13.   }
  14.   if (curr.size(curr.shape().length - 2) != curr.size(curr.shape().length - 1)) {
  15.     throw new AssertionError("Can only perform this operation on a SimpleTensor with square slices");
  16.   }
  17.   INDArray ret = Nd4j.create(curr.slices(), 1);
  18.   INDArray inT = in.transpose();
  19.   for (int i = 0; i < curr.slices(); i++) {
  20.     INDArray slice = curr.slice(i);
  21.     INDArray inTTimesSlice = inT.mmul(slice);
  22.     ret.putScalar(i, Nd4j.getBlasWrapper().dot(inTTimesSlice, in));
  23.   }
  24.   return ret;
  25. }

代码示例来源：origin: deeplearning4j/nd4j

  1. Nd4j.getBlasWrapper().axpy(Nd4j.UNIT, cOther, cResult);
  2. } else if (result == other) {
  3.   Nd4j.getBlasWrapper().axpy(Nd4j.UNIT, this, cResult);
  4. } else {
  5.   INDArray resultLinear = result.linearView();

代码示例来源：origin: deeplearning4j/nd4j

  1. Nd4j.getBlasWrapper().level2().gemv(BlasBufferUtil.getCharForTranspose(temp),
  2.           BlasBufferUtil.getCharForTranspose(this), Nd4j.UNIT, this, otherArray, Nd4j.ZERO, temp);
  3. } else {
  4.   Nd4j.getBlasWrapper().level3().gemm(BlasBufferUtil.getCharForTranspose(temp),
  5.           BlasBufferUtil.getCharForTranspose(this), BlasBufferUtil.getCharForTranspose(other),
  6.           Nd4j.UNIT, this, otherArray, Nd4j.ZERO, temp);
  7. Nd4j.getBlasWrapper().copy(temp, resultArray);
  8.   Nd4j.getBlasWrapper().level2().gemv(BlasBufferUtil.getCharForTranspose(resultArray),
  9.           BlasBufferUtil.getCharForTranspose(this), Nd4j.UNIT, this, otherArray, Nd4j.ZERO,
  10.           resultArray);
  11.   Nd4j.getBlasWrapper().level3().gemm(BlasBufferUtil.getCharForTranspose(resultArray),
  12.           BlasBufferUtil.getCharForTranspose(this), BlasBufferUtil.getCharForTranspose(other),
  13.           Nd4j.UNIT, this, otherArray, Nd4j.ZERO, resultArray);

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Compute generalized eigenvalues of the problem A x = L B x.
  3.  * The data will be unchanged, no eigenvectors returned unless calculateVectors is true.
  4.  * If calculateVectors == true, A will contain a matrix with the eigenvectors as columns.
  5.  *
  6.  * @param A symmetric Matrix A.
  7.  * @param B symmetric Matrix B.
  8.  * @return a vector of eigenvalues L.
  9.  */
  10. public static INDArray symmetricGeneralizedEigenvalues(INDArray A, INDArray B, boolean calculateVectors) {
  11.   assert A.rows() == A.columns();
  12.   assert B.rows() == B.columns();
  13.   INDArray W = Nd4j.create(A.rows());
  14.   if (calculateVectors)
  15.     A.assign(InvertMatrix.invert(B, false).mmuli(A));
  16.   else
  17.     A = InvertMatrix.invert(B, false).mmuli(A);
  18.   Nd4j.getBlasWrapper().syev('V', 'L', A, W);
  19.   return W;
  20. }

代码示例来源：origin: deeplearning4j/nd4j

  1. Nd4j.getBlasWrapper().lapack().gesvd(A, s, null, VT);

代码示例来源：origin: deeplearning4j/nd4j

  1. Nd4j.getBlasWrapper().level2().gemv(ordering(), BlasBufferUtil.getCharForTranspose(other), 1.0, this, other,
  2.           0.0, gemmResultArr);
  3. } else {
  4.   Nd4j.getBlasWrapper().level3().gemm(ordering(), BlasBufferUtil.getCharForTranspose(other),
  5.           BlasBufferUtil.getCharForTranspose(gemmResultArr), 1.0, this, other, 0.0, gemmResultArr);

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2. public INDArray sample(int[] shape) {
  3.   int numRows = 1;
  4.   for (int i = 0; i < shape.length - 1; i++)
  5.     numRows *= shape[i];
  6.   int numCols = shape[shape.length - 1];
  7.   val flatShape = new int[]{numRows, numCols};
  8.   val flatRng =  Nd4j.getExecutioner().exec(new GaussianDistribution(Nd4j.createUninitialized(flatShape, Nd4j.order()), 0.0, 1.0), random);
  9.   long m = flatRng.rows();
  10.   long n = flatRng.columns();
  11.   val s = Nd4j.create(m < n ? m : n);
  12.   val u = m < n ? Nd4j.create(m, n) : Nd4j.create(m, m);
  13.   val v = Nd4j.create(n, n, 'f');
  14.   Nd4j.getBlasWrapper().lapack().gesvd(flatRng, s, u, v);
  15.   // FIXME: int cast
  16.   if (gains == null) {
  17.     if (u.rows() == numRows && u.columns() == numCols) {
  18.       return v.get(NDArrayIndex.interval(0, numRows), NDArrayIndex.interval(0, numCols)).mul(gain).reshape(ArrayUtil.toLongArray(shape));
  19.     } else {
  20.       return u.get(NDArrayIndex.interval(0, numRows), NDArrayIndex.interval(0, numCols)).mul(gain).reshape(ArrayUtil.toLongArray(shape));
  21.     }
  22.   } else {
  23.     throw new UnsupportedOperationException();
  24.   }
  25. }

代码示例来源：origin: deeplearning4j/nd4j

  1. INDArray VL = Nd4j.create(A.rows(), A.rows());
  2. Nd4j.getBlasWrapper().geev('v', 'v', A.dup(), WR, WI, VL, VR);

内容来源于网络，如有侵权，请联系作者删除！

Nd4j

相关文章

    查看更多

    热门标签

    更多
    JavaquerypythonNode开发语言requestUtil数据库Table后端算法LoggerMessageElementParser

    最新文章

    更多

    Nd4j类方法

    • 蜀ICP备13028337号-1 大数据知识库 https://www.saoniuhuo.com © All rights reserved
    • 本站内容来源互联网,如果侵犯您的权益请联系我们删除, 联系方式：448109455@qq.com