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

x33g5p2x  于2022-01-24 转载在 其他  
字(9.7k)|赞(0)|评价(0)|浏览(219)

本文整理了Java中org.nd4j.linalg.factory.Nd4j.zeros()方法的一些代码示例，展示了Nd4j.zeros()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.zeros()方法的具体详情如下：
包路径：org.nd4j.linalg.factory.Nd4j
类名称：Nd4j
方法名：zeros

Nd4j.zeros介绍

[英]Creates a row vector with the specified number of columns
[中]创建具有指定列数的行向量

代码示例

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

  1. private static INDArray createFromCSC(double[] data, int[] rowIndices, int[] columnPointers, int[] shape){
  2.     INDArray result = Nd4j.zeros(shape);
  3.     int columns = shape[1];
  4.     int dataIdx = 0;
  5.     for(int i = 0; i < columns; i++){
  6.       for(int k = dataIdx; k < (i == columnPointers.length-1 ? rowIndices.length : columnPointers[i+1]); k++, dataIdx++){
  7.         int j = rowIndices[k];
  8.         result.put(j, i, data[k]);
  9.         //System.out.println("i = "+i+", k = "+k+ ", data[k] = "+data[k]+"\n matrix = "+result.toString());
  10.       }
  11.     }
  12.     return result;
  13.   }
  14. }

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

  1. @Override
  2.   public DataSet call(String s) throws Exception {
  3.     //Here: take a String, and map the characters to a one-hot representation
  4.     Map<Character, Integer> cti = ctiBroadcast.getValue();
  5.     int length = s.length();
  6.     INDArray features = Nd4j.zeros(1, N_CHARS, length - 1);
  7.     INDArray labels = Nd4j.zeros(1, N_CHARS, length - 1);
  8.     char[] chars = s.toCharArray();
  9.     int[] f = new int[3];
  10.     int[] l = new int[3];
  11.     for (int i = 0; i < chars.length - 2; i++) {
  12.       f[1] = cti.get(chars[i]);
  13.       f[2] = i;
  14.       l[1] = cti.get(chars[i + 1]);   //Predict the next character given past and current characters
  15.       l[2] = i;
  16.       features.putScalar(f, 1.0);
  17.       labels.putScalar(l, 1.0);
  18.     }
  19.     return new DataSet(features, labels);
  20.   }
  21. }

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

  1. INDArray zeros = Nd4j.zeros(nRows, nColumns);

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

  1. public DataSet convertDataSet(int num)  {
  2.   int batchNumCount = 0;
  3.   List<DataSet> dataSets = new ArrayList();
  4.   FileSystem fs = CommonUtils.openHdfsConnect();
  5.   try {
  6.     while (batchNumCount != num && fileIterator.hasNext()) {
  7.       ++ batchNumCount;
  8.       String fullPath = fileIterator.next();
  9.       Writable labelText = new Text(FilenameUtils.getBaseName((new File(fullPath)).getParent()));
  10.       INDArray features = null;
  11.       INDArray label =  Nd4j.zeros(1, labels.size()).putScalar(new int[]{0, labels.indexOf(labelText)}, 1);
  12.       InputStream imageios = fs.open(new Path(fullPath));
  13.       features = asMatrix(imageios);
  14.       imageios.close();
  15.       Nd4j.getAffinityManager().tagLocation(features, AffinityManager.Location.HOST);
  16.       dataSets.add(new DataSet(features, label));
  17.     }
  18.   } catch (Exception e) {
  19.     throw  new RuntimeException(e.getCause());
  20.   } finally {
  21.     CommonUtils.closeHdfsConnect(fs);
  22.   }
  23.   if (dataSets.size() == 0) {
  24.     return new DataSet();
  25.   } else {
  26.     DataSet result = DataSet.merge( dataSets );
  27.     return result;
  28.   }
  29. }

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

  1. INDArray initializationInput = Nd4j.zeros(numSamples, intToChar.size(), initialization.length());
  2. char[] init = initialization.toCharArray();
  3. for (int i = 0; i < init.length; i++) {
  4.   INDArray nextInput = Nd4j.zeros(numSamples, intToChar.size());

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

  1. INDArray featuresMask = Nd4j.zeros(reviews.size(), maxLength);
  2. INDArray labelsMask = Nd4j.zeros(reviews.size(), maxLength);

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

  1. INDArray myArray = Nd4j.zeros(nRows, nColumns);

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

  1. INDArray zerosColumn = Nd4j.zeros(3,1);
  2. originalArray.put(new INDArrayIndex[]{NDArrayIndex.all(), NDArrayIndex.point(2)}, zerosColumn);     //All rows, column index 2
  3. System.out.println("\n\n\nOriginal array, after put operation:\n" + originalArray);

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

  1. INDArray initializationInput = Nd4j.zeros(numSamples, iter.inputColumns(), initialization.length());
  2. char[] init = initialization.toCharArray();
  3. for( int i=0; i<init.length; i++ ){
  4.   INDArray nextInput = Nd4j.zeros(numSamples,iter.inputColumns());

代码示例来源：origin: deeplearning4j/nd4j

  1. @Override
  2. public INDArray toDense() {
  3.   // Dummy way - going to use the conversion routines in level2 (?)
  4.   INDArray result = Nd4j.zeros(shape());
  5.   int[] pointersB = pointerB.asInt();
  6.   int[] pointersE = pointerE.asInt();
  7.   for (int row = 0; row < rows(); row++) {
  8.     for (int idx = pointersB[row]; idx < pointersE[row]; idx++) {
  9.       result.put(row, columnsPointers.getInt(idx), values.getNumber(idx));
  10.     }
  11.   }
  12.   return result;
  13. }

代码示例来源：origin: guoguibing/librec

  1. @Override
  2. protected void setup() throws LibrecException {
  3.   super.setup();
  4.   inputDim = numUsers;
  5.   hiddenDim = conf.getInt("rec.hidden.dimension");
  6.   learningRate = conf.getDouble("rec.iterator.learnrate");
  7.   lambdaReg = conf.getDouble("rec.weight.regularization");
  8.   numIterations = conf.getInt("rec.iterator.maximum");
  9.   hiddenActivation = conf.get("rec.hidden.activation");
  10.   outputActivation = conf.get("rec.output.activation");
  11.   // transform the sparse matrix to INDArray
  12.   int[] matrixShape = {numItems, numUsers};
  13.   trainSet = Nd4j.zeros(matrixShape);
  14.   trainSetMask = Nd4j.zeros(matrixShape);
  15.   for (MatrixEntry me: trainMatrix) {
  16.     trainSet.put(me.column(), me.row(), me.get());
  17.     trainSetMask.put(me.column(), me.row(), 1);
  18.   }
  19. }

代码示例来源：origin: deeplearning4j/nd4j

  1. private INDArray labelsMinusMu(INDArray labels, INDArray mu) {
  2.   // Now that we have the mixtures, let's compute the negative
  3.   // log likelihodd of the label against the 
  4.   long nSamples = labels.size(0);
  5.   long labelsPerSample = labels.size(1);
  6.   // This worked, but was actually much
  7.   // slower than the for loop below.
  8.   // labels = samples, mixtures, labels
  9.   // mu = samples, mixtures
  10.   // INDArray labelMinusMu = labels
  11.   //        .reshape('f', nSamples, labelsPerSample, 1)
  12.   //        .repeat(2, mMixtures)
  13.   //        .permute(0, 2, 1)
  14.   //        .subi(mu);
  15.   // The above code does the same thing as the loop below,
  16.   // but it does it with index magix instead of a for loop.
  17.   // It turned out to be way less efficient than the simple 'for' here.
  18.   INDArray labelMinusMu = Nd4j.zeros(nSamples, mMixtures, labelsPerSample);
  19.   for (int k = 0; k < mMixtures; k++) {
  20.     labelMinusMu.put(new INDArrayIndex[] {NDArrayIndex.all(), NDArrayIndex.point(k), NDArrayIndex.all()},
  21.             labels);
  22.   }
  23.   labelMinusMu.subi(mu);
  24.   return labelMinusMu;
  25. }

代码示例来源：origin: deeplearning4j/nd4j

  1. /**
  2.  * Converts the sparse ndarray into a dense one
  3.  * @return a dense ndarray
  4.  */
  5. @Override
  6. public INDArray toDense() {
  7.   // TODO support view conversion
  8.   INDArray result = Nd4j.zeros(shape());
  9.   switch (data().dataType()) {
  10.     case DOUBLE:
  11.       for (int i = 0; i < length; i++) {
  12.         int[] idx = getUnderlyingIndicesOf(i).asInt();
  13.         double value = values.getDouble(i);
  14.         result.putScalar(idx, value);
  15.       }
  16.       break;
  17.     case FLOAT:
  18.       for (int i = 0; i < length; i++) {
  19.         int[] idx = getUnderlyingIndicesOf(i).asInt();
  20.         float value = values.getFloat(i);
  21.         result.putScalar(idx, value);
  22.       }
  23.       break;
  24.     default:
  25.       throw new UnsupportedOperationException();
  26.   }
  27.   return result;
  28. }

代码示例来源：origin: guoguibing/librec

  1. @Override
  2. protected void setup() throws LibrecException {
  3.   super.setup();
  4.   inputDim = numItems;
  5.   hiddenDim = conf.getInt("rec.hidden.dimension");
  6.   learningRate = conf.getDouble("rec.iterator.learnrate");
  7.   lambdaReg = conf.getDouble("rec.weight.regularization");
  8.   numIterations = conf.getInt("rec.iterator.maximum");
  9.   hiddenActivation = conf.get("rec.hidden.activation");
  10.   outputActivation = conf.get("rec.output.activation");
  11.   // transform the sparse matrix to INDArray
  12.   // the sparse training matrix has been binarized
  13.   int[] matrixShape = {numUsers, numItems};
  14.   trainSet = Nd4j.zeros(matrixShape);
  15.   for (MatrixEntry me: trainMatrix) {
  16.     trainSet.put(me.row(), me.column(), me.get());
  17.   }
  18. }

代码示例来源：origin: deeplearning4j/nd4j

  1. public static boolean checkMulManually(INDArray first, INDArray second, double maxRelativeDifference,
  2.         double minAbsDifference) {
  3.   //No apache commons element-wise multiply, but can do this manually
  4.   INDArray result = first.mul(second);
  5.   long[] shape = first.shape();
  6.   INDArray expected = Nd4j.zeros(first.shape());
  7.   for (int i = 0; i < shape[0]; i++) {
  8.     for (int j = 0; j < shape[1]; j++) {
  9.       double v = first.getDouble(i, j) * second.getDouble(i, j);
  10.       expected.putScalar(new int[] {i, j}, v);
  11.     }
  12.   }
  13.   if (!checkShape(expected, result))
  14.     return false;
  15.   boolean ok = checkEntries(expected, result, maxRelativeDifference, minAbsDifference);
  16.   if (!ok) {
  17.     INDArray onCopies = Shape.toOffsetZeroCopy(first).mul(Shape.toOffsetZeroCopy(second));
  18.     printFailureDetails(first, second, expected, result, onCopies, "mul");
  19.   }
  20.   return ok;
  21. }

代码示例来源：origin: deeplearning4j/nd4j

  1. public static boolean checkDivManually(INDArray first, INDArray second, double maxRelativeDifference,
  2.         double minAbsDifference) {
  3.   //No apache commons element-wise division, but can do this manually
  4.   INDArray result = first.div(second);
  5.   long[] shape = first.shape();
  6.   INDArray expected = Nd4j.zeros(first.shape());
  7.   for (int i = 0; i < shape[0]; i++) {
  8.     for (int j = 0; j < shape[1]; j++) {
  9.       double v = first.getDouble(i, j) / second.getDouble(i, j);
  10.       expected.putScalar(new int[] {i, j}, v);
  11.     }
  12.   }
  13.   if (!checkShape(expected, result))
  14.     return false;
  15.   boolean ok = checkEntries(expected, result, maxRelativeDifference, minAbsDifference);
  16.   if (!ok) {
  17.     INDArray onCopies = Shape.toOffsetZeroCopy(first).mul(Shape.toOffsetZeroCopy(second));
  18.     printFailureDetails(first, second, expected, result, onCopies, "div");
  19.   }
  20.   return ok;
  21. }

代码示例来源：origin: deeplearning4j/nd4j

  1. public INDArray getGradient(INDArray gradient, int slice, int[] shape) {
  2.   boolean historicalInitialized = false;
  3.   INDArray sqrtHistory;
  4.   if (this.historicalGradient == null) {
  5.     this.historicalGradient = Nd4j.zeros(shape).add(epsilon);
  6.     historicalInitialized = true;
  7.   } else if (!this.historicalGradient.isVector()
  8.           && this.historicalGradient.slice(slice).length() != gradient.length())
  9.     throw new IllegalArgumentException("Illegal gradient");
  10.   if (historicalGradient.isVector())
  11.     sqrtHistory = sqrt(historicalGradient);
  12.   else
  13.     sqrtHistory = !historicalInitialized ? sqrt(historicalGradient.slice(slice)) : historicalGradient;
  14.   INDArray learningRates;
  15.   try {
  16.     learningRates = sqrtHistory.rdivi(learningRate);
  17.   } catch (ArithmeticException ae) {
  18.     learningRates = sqrtHistory.rdivi(learningRate + epsilon);
  19.   }
  20.   if (gradient.length() != learningRates.length())
  21.     gradient.muli(learningRates.slice(slice));
  22.   else
  23.     gradient.muli(learningRates);
  24.   this.historicalGradient.slice(slice).addi(gradient.mul(gradient));
  25.   numIterations++;
  26.   //ensure no zeros
  27.   return gradient;
  28. }

代码示例来源：origin: deeplearning4j/nd4j

  1. INDArray bernoullis = Nd4j.zeros(labelMask.shape());
  2. long currentTimeSliceEnd = label.size(2);

代码示例来源：origin: deeplearning4j/nd4j

  1. std = (batchCount == 1) ? Nd4j.zeros(mean.shape()) : Transforms.pow(next.getFeatureMatrix().std(0), 2);
  2.   std.muli(batchCount);
  3. } else {

代码示例来源：origin: deeplearning4j/nd4j

  1. INDArray gradient = Nd4j.zeros(nSamples, preOutput.columns());

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