如何访问 tf.NamedTensorMap 中的数据？

所以，我似乎到处寻找一些文档来解释tf.NamedTensorMap，但它似乎并不存在。在玩弄了Object类的方法之后，我意识到我可以使用Object.getPrototypeOf()来访问所有的方法名称。此外，通过使用Object.keys()，我能够控制台记录tf.NamedTensorMap中的所有属性键。所以这就是我发现的。

似乎tf.NamedTensorMap.indices、tf.NamedTensorMap.values和tf.NamedTensorMap.shape中的每一个由于某种原因具有相同的方法，即使它们持有不同的数据。我假设这是由于一些继承。无论如何，我能够找到多种可用于从张量获取数据的方法。即tf.dataSync()和tf.arraySync()两者都适用于我的特定场景。

Tensor {
  rank: [Getter],
  buffer: [Function (anonymous)],
  bufferSync: [Function (anonymous)],
  array: [Function (anonymous)],
  arraySync: [Function (anonymous)],
  data: [Function (anonymous)],
  dataToGPU: [Function (anonymous)],
  dataSync: [Function (anonymous)],
  bytes: [Function (anonymous)],
  dispose: [Function (anonymous)],
  isDisposed: [Getter],
  throwIfDisposed: [Function (anonymous)],
  print: [Function (anonymous)],
  clone: [Function (anonymous)],
  toString: [Function (anonymous)],
  cast: [Function (anonymous)],
  variable: [Function (anonymous)],
  abs: [Function (anonymous)],
  acos: [Function (anonymous)],
  acosh: [Function (anonymous)],
  add: [Function (anonymous)],
  all: [Function (anonymous)],
  any: [Function (anonymous)],
  argMax: [Function (anonymous)],
  argMin: [Function (anonymous)],
  asScalar: [Function (anonymous)],
  asType: [Function (anonymous)],
  as1D: [Function (anonymous)],
  as2D: [Function (anonymous)],
  as3D: [Function (anonymous)],
  as4D: [Function (anonymous)],
  as5D: [Function (anonymous)],
  asin: [Function (anonymous)],
  asinh: [Function (anonymous)],
  atan: [Function (anonymous)],
  atan2: [Function (anonymous)],
  atanh: [Function (anonymous)],
  avgPool: [Function (anonymous)],
  batchToSpaceND: [Function (anonymous)],
  batchNorm: [Function (anonymous)],
  broadcastTo: [Function (anonymous)],
  ceil: [Function (anonymous)],
  clipByValue: [Function (anonymous)],
  concat: [Function (anonymous)],
  conv1d: [Function (anonymous)],
  conv2dTranspose: [Function (anonymous)],
  conv2d: [Function (anonymous)],
  cos: [Function (anonymous)],
  cosh: [Function (anonymous)],
  cumprod: [Function (anonymous)],
  cumsum: [Function (anonymous)],
  depthToSpace: [Function (anonymous)],
  depthwiseConv2d: [Function (anonymous)],
  dilation2d: [Function (anonymous)],
  divNoNan: [Function (anonymous)],
  div: [Function (anonymous)],
  dot: [Function (anonymous)],
  elu: [Function (anonymous)],
  equal: [Function (anonymous)],
  erf: [Function (anonymous)],
  exp: [Function (anonymous)],
  expandDims: [Function (anonymous)],
  expm1: [Function (anonymous)],
  fft: [Function (anonymous)],
  flatten: [Function (anonymous)],
  floor: [Function (anonymous)],
  floorDiv: [Function (anonymous)],
  gather: [Function (anonymous)],
  greaterEqual: [Function (anonymous)],
  greater: [Function (anonymous)],
  ifft: [Function (anonymous)],
  irfft: [Function (anonymous)],
  isFinite: [Function (anonymous)],
  isInf: [Function (anonymous)],
  isNaN: [Function (anonymous)],
  leakyRelu: [Function (anonymous)],
  lessEqual: [Function (anonymous)],
  less: [Function (anonymous)],
  localResponseNormalization: [Function (anonymous)],
  logSigmoid: [Function (anonymous)],
  logSoftmax: [Function (anonymous)],
  logSumExp: [Function (anonymous)],
  log: [Function (anonymous)],
  log1p: [Function (anonymous)],
  logicalAnd: [Function (anonymous)],
  logicalNot: [Function (anonymous)],
  logicalOr: [Function (anonymous)],
  logicalXor: [Function (anonymous)],
  matMul: [Function (anonymous)],
  maxPool: [Function (anonymous)],
  max: [Function (anonymous)],
  maximum: [Function (anonymous)],
  mean: [Function (anonymous)],
  min: [Function (anonymous)],
  minimum: [Function (anonymous)],
  mirrorPad: [Function (anonymous)],
  mod: [Function (anonymous)],
  mul: [Function (anonymous)],
  neg: [Function (anonymous)],
  norm: [Function (anonymous)],
  notEqual: [Function (anonymous)],
  oneHot: [Function (anonymous)],
  onesLike: [Function (anonymous)],
  pad: [Function (anonymous)],
  pool: [Function (anonymous)],
  pow: [Function (anonymous)],
  prelu: [Function (anonymous)],
  prod: [Function (anonymous)],
  reciprocal: [Function (anonymous)],
  relu: [Function (anonymous)],
  relu6: [Function (anonymous)],
  reshapeAs: [Function (anonymous)],
  reshape: [Function (anonymous)],
  resizeBilinear: [Function (anonymous)],
  resizeNearestNeighbor: [Function (anonymous)],
  reverse: [Function (anonymous)],
  rfft: [Function (anonymous)],
  round: [Function (anonymous)],
  rsqrt: [Function (anonymous)],
  selu: [Function (anonymous)],
  separableConv2d: [Function (anonymous)],
  sigmoid: [Function (anonymous)],
  sign: [Function (anonymous)],
  sin: [Function (anonymous)],
  sinh: [Function (anonymous)],
  slice: [Function (anonymous)],
  softmax: [Function (anonymous)],
  softplus: [Function (anonymous)],
  spaceToBatchND: [Function (anonymous)],
  split: [Function (anonymous)],
  sqrt: [Function (anonymous)],
  square: [Function (anonymous)],
  squaredDifference: [Function (anonymous)],
  squeeze: [Function (anonymous)],
  stack: [Function (anonymous)],
  step: [Function (anonymous)],
  stridedSlice: [Function (anonymous)],
  sub: [Function (anonymous)],
  sum: [Function (anonymous)],
  tan: [Function (anonymous)],
  tanh: [Function (anonymous)],
  tile: [Function (anonymous)],
  toBool: [Function (anonymous)],
  toFloat: [Function (anonymous)],
  toInt: [Function (anonymous)],
  topk: [Function (anonymous)],
  transpose: [Function (anonymous)],
  unique: [Function (anonymous)],
  unsortedSegmentSum: [Function (anonymous)],
  unstack: [Function (anonymous)],
  where: [Function (anonymous)],
  zerosLike: [Function (anonymous)]
}

由于张量是如何工作的，如果我现在理解它们，尝试制作一个包含张量的张量是一个坏主意，它甚至可能不起作用（我必须对其进行测试）。因此，如果您的数据与我正在使用的数据类似，并且您需要在数据集中的较低级别拆分数据，那么使用基本的string.split()可能会更好。