Python array_ops.tile函数代码示例

def _mask_probs(probs, eos_token, finished):
  """Masks log probabilities.

  The result is that finished beams allocate all probability mass to eos and
  unfinished beams remain unchanged.

  Args:
    probs: Log probabiltiies of shape `[batch_size, beam_width, vocab_size]`
    eos_token: An int32 id corresponding to the EOS token to allocate
      probability to.
    finished: A boolean tensor of shape `[batch_size, beam_width]` that
      specifies which elements in the beam are finished already.

  Returns:
    A tensor of shape `[batch_size, beam_width, vocab_size]`, where unfinished
    beams stay unchanged and finished beams are replaced with a tensor with all
    probability on the EOS token.
  """
  vocab_size = array_ops.shape(probs)[2]
  # All finished examples are replaced with a vector that has all
  # probability on EOS
  finished_row = array_ops.one_hot(
      eos_token,
      vocab_size,
      dtype=probs.dtype,
      on_value=0.,
      off_value=probs.dtype.min)
  finished_probs = array_ops.tile(
      array_ops.reshape(finished_row, [1, 1, -1]),
      array_ops.concat([array_ops.shape(finished), [1]], 0))
  finished_mask = array_ops.tile(
      array_ops.expand_dims(finished, 2), [1, 1, vocab_size])

  return array_ops.where(finished_mask, finished_probs, probs)

def _SumGrad(op, grad):
  """Gradient for Sum."""
  # Fast path for when reducing to a scalar and ndims is known: adds only
  # Reshape and Tile ops (and possibly a Shape).
  input_0_shape = op.inputs[0]._shape_tuple()  # pylint: disable=protected-access
  if input_0_shape is not None:
    axes = tensor_util.constant_value(op.inputs[1])
    if axes is not None:
      rank = len(input_0_shape)
      if np.array_equal(axes, np.arange(rank)):  # Reduce all dims.
        grad = array_ops.reshape(grad, [1] * rank)
        # If shape is not fully defined (but rank is), we use Shape.
        if None not in input_0_shape:
          input_shape = input_0_shape
        else:
          input_shape = array_ops.shape(op.inputs[0])
        return [array_ops.tile(grad, input_shape), None]

  input_shape = array_ops.shape(op.inputs[0])
  # TODO(apassos) remove this once device placement for eager ops makes more
  # sense.
  with ops.colocate_with(input_shape):
    output_shape_kept_dims = math_ops.reduced_shape(input_shape, op.inputs[1])
    tile_scaling = _safe_shape_div(input_shape, output_shape_kept_dims)
  grad = array_ops.reshape(grad, output_shape_kept_dims)
  return [array_ops.tile(grad, tile_scaling), None]

  def testShapeFunctionEdgeCases(self):
    # Unknown multiples shape.
    inp = constant_op.constant(0.0, shape=[4, 4, 4, 4])
    tiled = array_ops.tile(inp, array_ops.placeholder(dtypes.int32))
    self.assertEqual([None, None, None, None], tiled.get_shape().as_list())

    # Unknown input shape.
    inp = array_ops.placeholder(dtypes.float32)
    tiled = array_ops.tile(inp, [2, 2, 2, 2])
    self.assertEqual([None, None, None, None], tiled.get_shape().as_list())

    # Unknown input and multiples shape.
    inp = array_ops.placeholder(dtypes.float32)
    tiled = array_ops.tile(inp, array_ops.placeholder(dtypes.int32))
    self.assertIs(None, tiled.get_shape().ndims)

    # Known input and partially known multiples.
    inp = constant_op.constant(0.0, shape=[1, 1])
    tiled = array_ops.tile(inp, [array_ops.placeholder(dtypes.int32), 7])
    self.assertEqual([None, 7], tiled.get_shape().as_list())

    # Mismatched input rank and multiples length.
    inp = array_ops.placeholder(dtypes.float32, shape=[None, None])
    with self.assertRaises(ValueError):
      tiled = array_ops.tile(
          inp, array_ops.placeholder(
              dtypes.int32, shape=[3]))

def frames(signal, frame_length, frame_step, name=None):
  """Frame a signal into overlapping frames.

  May be used in front of spectral functions.

  For example:

  ```python
  pcm = tf.placeholder(tf.float32, [None, 9152])
  frames = tf.contrib.signal.frames(pcm, 512, 180)
  magspec = tf.abs(tf.spectral.rfft(frames, [512]))
  image = tf.expand_dims(magspec, 3)
  ```

  Args:
    signal: A `Tensor` of shape `[batch_size, signal_length]`.
    frame_length: An `int32` or `int64` `Tensor`. The length of each frame.
    frame_step: An `int32` or `int64` `Tensor`. The step between frames.
    name: A name for the operation (optional).

  Returns:
    A `Tensor` of frames with shape `[batch_size, num_frames, frame_length]`.

  Raises:
    ValueError: if signal does not have rank 2.
  """
  with ops.name_scope(name, "frames", [signal, frame_length, frame_step]):
    signal = ops.convert_to_tensor(signal, name="signal")
    frame_length = ops.convert_to_tensor(frame_length, name="frame_length")
    frame_step = ops.convert_to_tensor(frame_step, name="frame_step")

    signal_rank = signal.shape.ndims

    if signal_rank != 2:
      raise ValueError("expected signal to have rank 2 but was " + signal_rank)

    signal_length = array_ops.shape(signal)[1]

    num_frames = math_ops.ceil((signal_length - frame_length) / frame_step)
    num_frames = 1 + math_ops.cast(num_frames, dtypes.int32)

    pad_length = (num_frames - 1) * frame_step + frame_length
    pad_signal = array_ops.pad(signal, [[0, 0], [0,
                                                 pad_length - signal_length]])

    indices_frame = array_ops.expand_dims(math_ops.range(frame_length), 0)
    indices_frames = array_ops.tile(indices_frame, [num_frames, 1])

    indices_step = array_ops.expand_dims(
        math_ops.range(num_frames) * frame_step, 1)
    indices_steps = array_ops.tile(indices_step, [1, frame_length])

    indices = indices_frames + indices_steps

    # TODO(androbin): remove `transpose` when `gather` gets `axis` support
    pad_signal = array_ops.transpose(pad_signal)
    signal_frames = array_ops.gather(pad_signal, indices)
    signal_frames = array_ops.transpose(signal_frames, perm=[2, 0, 1])

    return signal_frames

  def to_weighted_sum(self,
                      input_tensor,
                      num_outputs=1,
                      weight_collections=None,
                      trainable=True):
    """Returns a Tensor as linear predictions and a list of created Variable."""
    dimension = self.source_column.dimension
    batch_size = array_ops.shape(input_tensor)[0]

    if dimension > 1:
      i1 = array_ops.reshape(array_ops.tile(array_ops.expand_dims(
          math_ops.range(0, batch_size), 1), [1, dimension]), [-1])
      i2 = array_ops.tile(math_ops.range(0, dimension), [batch_size])
      # Flatten the bucket indices and unique them across dimensions
      # E.g. 2nd dimension indices will range from k to 2*k-1 with k buckets
      # TODO(chapelle): move that logic to insert_transformed_feature to ensure
      #   unique buckets across dimensions after crossing.
      bucket_indices = array_ops.reshape(input_tensor, [-1]) + self.length * i2
    else:
      # Simpler indices when dimension=1
      i1 = math_ops.range(0, batch_size)
      i2 = array_ops.zeros([batch_size], dtype=dtypes.int32)
      bucket_indices = array_ops.reshape(input_tensor, [-1])

    indices = math_ops.to_int64(array_ops.transpose(array_ops.pack((i1, i2))))
    shape = math_ops.to_int64(array_ops.pack([batch_size, 1]))
    sparse_id_values = ops.SparseTensor(indices, bucket_indices, shape)
    vocab_size = self.length * self.source_column.dimension

    return _create_embedding_lookup(
        sparse_id_values, vocab_size, num_outputs,
        _add_variable_collection(weight_collections), 0., "sum",
        trainable, self.name + "_weights")

def gather_tree_from_array(t, parent_ids, sequence_length):
  """Calculates the full beams for `TensorArray`s.

  Args:
    t: A stacked `TensorArray` of size `max_time` that contains `Tensor`s of
      shape `[batch_size, beam_width, s]` or `[batch_size * beam_width, s]`
      where `s` is the depth shape.
    parent_ids: The parent ids of shape `[max_time, batch_size, beam_width]`.
    sequence_length: The sequence length of shape `[batch_size, beam_width]`.

  Returns:
    A `Tensor` which is a stacked `TensorArray` of the same size and type as
    `t` and where beams are sorted in each `Tensor` according to `parent_ids`.
  """
  max_time = parent_ids.shape[0].value or array_ops.shape(parent_ids)[0]
  batch_size = parent_ids.shape[1].value or array_ops.shape(parent_ids)[1]
  beam_width = parent_ids.shape[2].value or array_ops.shape(parent_ids)[2]

  # Generate beam ids that will be reordered by gather_tree.
  beam_ids = array_ops.expand_dims(
      array_ops.expand_dims(math_ops.range(beam_width), 0), 0)
  beam_ids = array_ops.tile(beam_ids, [max_time, batch_size, 1])

  mask = array_ops.sequence_mask(
      sequence_length, maxlen=max_time, dtype=dtypes.int32)
  mask = array_ops.transpose(mask, perm=[2, 0, 1])

  # Use beam_width + 1 to mark the end of beam.
  masked_beam_ids = (beam_ids * mask) + (1 - mask) * (beam_width + 1)

  max_sequence_lengths = math_ops.to_int32(
      math_ops.reduce_max(sequence_length, axis=1))
  sorted_beam_ids = beam_search_ops.gather_tree(
      step_ids=masked_beam_ids,
      parent_ids=parent_ids,
      max_sequence_lengths=max_sequence_lengths,
      end_token=beam_width + 1)

  # For out of range steps, simply copy the same beam.
  sorted_beam_ids = array_ops.where(
      math_ops.cast(mask, dtypes.bool), x=sorted_beam_ids, y=beam_ids)

  # Generate indices for gather_nd.
  time_ind = array_ops.tile(array_ops.reshape(
      math_ops.range(max_time), [-1, 1, 1]), [1, batch_size, beam_width])
  batch_ind = array_ops.tile(array_ops.reshape(
      math_ops.range(batch_size), [-1, 1, 1]), [1, max_time, beam_width])
  batch_ind = array_ops.transpose(batch_ind, perm=[1, 0, 2])
  indices = array_ops.stack([time_ind, batch_ind, sorted_beam_ids], -1)

  # Gather from a tensor with collapsed additional dimensions.
  gather_from = t
  final_shape = array_ops.shape(gather_from)
  gather_from = array_ops.reshape(
      gather_from, [max_time, batch_size, beam_width, -1])
  ordered = array_ops.gather_nd(gather_from, indices)
  ordered = array_ops.reshape(ordered, final_shape)

  return ordered

 def make_tril_ids(n):
   """Internal helper to create vector of linear indices into y."""
   cols = array_ops.reshape(array_ops.tile(math_ops.range(n), [n]), [n, n])
   rows = array_ops.tile(
       array_ops.expand_dims(math_ops.range(n), -1), [1, n])
   pred = math_ops.greater(cols, rows)
   tril_ids = array_ops.tile(array_ops.reshape(
       math_ops.cumsum(math_ops.range(n)), [n, 1]), [1, n]) + cols
   tril_ids = math_ops.select(pred,
                              array_ops.zeros([n, n], dtype=dtypes.int32),
                              tril_ids + 1)
   tril_ids = array_ops.reshape(tril_ids, [-1])
   return tril_ids

 def testInvalidDim(self):
   with self.test_session():
     inp = np.random.rand(4, 1).astype("f")
     a = constant_op.constant(
         [float(x) for x in inp.ravel(order="C")],
         shape=[4, 1],
         dtype=dtypes.float32)
     # Wrong length of multiples.
     with self.assertRaises(ValueError):
       array_ops.tile(a, [1, 4, 2])
     # Wrong rank for multiples.
     with self.assertRaises(ValueError):
       array_ops.tile(a, [[2, 3], [3, 4]]).eval()

  def _initialize_variables(self, data, initial_means=None):
    """Initializes variables.

    Args:
      data: a list of Tensors with data, each row is a new example.
      initial_means: a Tensor with a matrix of means.
    """
    first_shard = data[0]
    # Initialize means: num_classes X 1 X dimensions.
    if initial_means is not None:
      means = array_ops.expand_dims(initial_means, 1)
    else:
      # Sample data randomly
      means = array_ops.expand_dims(
          _init_clusters_random(data, self._num_classes, self._random_seed), 1)

    # Initialize covariances.
    if self._covariance_type == FULL_COVARIANCE:
      cov = _covariance(first_shard, False) + self._min_var
      # A matrix per class, num_classes X dimensions X dimensions
      covs = array_ops.tile(
          array_ops.expand_dims(cov, 0), [self._num_classes, 1, 1])
    elif self._covariance_type == DIAG_COVARIANCE:
      cov = _covariance(first_shard, True) + self._min_var
      # A diagonal per row, num_classes X dimensions.
      covs = array_ops.tile(
          array_ops.expand_dims(array_ops.diag_part(cov), 0),
          [self._num_classes, 1])

    with ops.colocate_with(self._cluster_centers_initialized):
      initialized = control_flow_ops.with_dependencies(
          [means, covs],
          array_ops.identity(self._cluster_centers_initialized))
    self._init_ops = []
    with ops.colocate_with(self._means):
      init_means = state_ops.assign(self._means, means, validate_shape=False)
      init_means = control_flow_ops.with_dependencies(
          [init_means],
          state_ops.assign(self._cluster_centers_initialized, True))
      self._init_ops.append(control_flow_ops.cond(initialized,
                                                  control_flow_ops.no_op,
                                                  lambda: init_means).op)
    with ops.colocate_with(self._covs):
      init_covs = state_ops.assign(self._covs, covs, validate_shape=False)
      init_covs = control_flow_ops.with_dependencies(
          [init_covs],
          state_ops.assign(self._cluster_centers_initialized, True))
      self._init_ops.append(control_flow_ops.cond(initialized,
                                                  control_flow_ops.no_op,
                                                  lambda: init_covs).op)

def _ApplyLengthsToBatch(sequence_lengths, tf_output):
  # TODO(drpng): just use Update so that we don't carry over the gradients?
  """Sets the output to be zero at the end of the sequence."""
  # output is batch major.
  batch_size, max_time, vector_size = tf_output.shape
  output_time = array_ops.tile(math_ops.range(0, max_time), [batch_size])
  output_time = array_ops.reshape(output_time, [batch_size, max_time])
  lengths = array_ops.tile(
      array_ops.reshape(sequence_lengths, [-1, 1]), [1, max_time])
  is_less = math_ops.cast(
      math_ops.less(output_time, lengths), dtype=dtypes.float32)
  keep_mask = array_ops.tile(
      array_ops.expand_dims(is_less, -1),
      [1, 1, vector_size])
  final_output = keep_mask * tf_output
  return final_output

  def inverse_stft_window_fn_inner(frame_length, dtype):
    """Computes a window that can be used in `inverse_stft`.

    Args:
      frame_length: An integer scalar `Tensor`. The window length in samples.
      dtype: Data type of waveform passed to `stft`.

    Returns:
      A window suitable for reconstructing original waveform in `inverse_stft`.

    Raises:
      ValueError: If `frame_length` is not scalar, `forward_window_fn` is not a
      callable that takes a window length and a `dtype` keyword argument and
      returns a `[window_length]` `Tensor` of samples in the provided datatype
      `frame_step` is not scalar, or `frame_step` is not scalar.
    """
    with ops.name_scope(name, 'inverse_stft_window_fn', [forward_window_fn]):
      frame_length = ops.convert_to_tensor(frame_length, name='frame_length')
      frame_length.shape.assert_has_rank(0)

      # Use equation 7 from Griffin + Lim.
      forward_window = forward_window_fn(frame_length, dtype=dtype)
      denom = math_ops.square(forward_window)
      overlaps = -(-frame_length // frame_step)  # Ceiling division.
      denom = array_ops.pad(denom, [(0, overlaps * frame_step - frame_length)])
      denom = array_ops.reshape(denom, [overlaps, frame_step])
      denom = math_ops.reduce_sum(denom, 0, keep_dims=True)
      denom = array_ops.tile(denom, [overlaps, 1])
      denom = array_ops.reshape(denom, [overlaps * frame_step])

      return forward_window / denom[:frame_length]

 def dataset_fn():
   dataset = dataset_ops.Dataset.range(10).map(
       lambda x: array_ops.tile([x], ops.convert_to_tensor([x])),
       num_parallel_calls=optimization.AUTOTUNE)
   options = dataset_ops.Options()
   options.experimental_autotune = True
   return dataset.with_options(options)

  def testPrefetchBufferUtilization(self, dataset_transformation):
    aggregator = stats_aggregator.StatsAggregator()
    dataset = dataset_ops.Dataset.range(100).map(
        lambda x: array_ops.tile([x], ops.convert_to_tensor([x]))).prefetch(-1)
    dataset = dataset_transformation(dataset, aggregator)
    iterator = dataset_ops.make_initializable_iterator(dataset)
    next_element = iterator.get_next()
    summary_t = aggregator.get_summary()

    with self.cached_session() as sess:
      self.evaluate(iterator.initializer)
      for i in range(100):
        self.assertAllEqual(
            np.array([i] * i, dtype=np.int64), self.evaluate(next_element))
        summary_str = self.evaluate(summary_t)
        self._assertSummaryHasCount(summary_str, "Prefetch::buffer_utilization",
                                    float(i + 1))
        self._assertSummaryContains(summary_str, "Prefetch::buffer_capacity")
        self._assertSummaryContains(summary_str, "Prefetch::buffer_size")
        self._assertSummaryHasRange(summary_str, "Prefetch::buffer_utilization",
                                    0, 1)
      with self.assertRaises(errors.OutOfRangeError):
        self.evaluate(next_element)
      summary_str = self.evaluate(summary_t)
      self._assertSummaryHasCount(summary_str, "Prefetch::buffer_utilization",
                                  100)

    def runFiniteDifferences(self, shapes, dtypes=(dtypes_lib.float32, dtypes_lib.float64), scalarTest=False):
        with self.test_session(use_gpu=False):
            for shape in shapes:
                for batch in False, True:
                    for dtype in dtypes:
                        if not scalarTest:
                            x = constant_op.constant(np.random.randn(shape[0], shape[1]), dtype)
                            tensor = math_ops.matmul(x, array_ops.transpose(x)) / shape[0]
                        else:
                            # This is designed to be a faster test for larger matrices.
                            x = constant_op.constant(np.random.randn(), dtype)
                            R = constant_op.constant(np.random.randn(shape[0], shape[1]), dtype)
                            e = math_ops.mul(R, x)
                            tensor = math_ops.matmul(e, array_ops.transpose(e)) / shape[0]

                        # Inner-most matrices in tensor are positive definite.
                        if batch:
                            tensor = array_ops.tile(array_ops.expand_dims(tensor, 0), [4, 1, 1])
                        y = linalg_ops.cholesky(tensor)
                        if scalarTest:
                            y = math_ops.reduce_mean(y)
                        error = gradient_checker.compute_gradient_error(x, x._shape_as_list(), y, y._shape_as_list())
                        tf_logging.info("error = %f", error)
                        if dtype == dtypes_lib.float64:
                            self.assertLess(error, 1e-5)
                        else:
                            self.assertLess(error, 3e-3)

 def testPrefetchBufferUtilization(self, dataset_transformation):
   aggregator = stats_aggregator.StatsAggregator()
   dataset = dataset_ops.Dataset.range(100).map(
       lambda x: array_ops.tile([x], ops.convert_to_tensor([x]))).prefetch(-1)
   dataset = dataset_transformation(dataset, aggregator)
   next_element = self.getNext(dataset, requires_initialization=True)
   for i in range(100):
     self.assertAllEqual(
         np.array([i] * i, dtype=np.int64), self.evaluate(next_element()))
     summary_str = self.evaluate(aggregator.get_summary())
     self._assertSummaryHasCount(
         summary_str,
         self.regexForNodeName("PrefetchDataset", "buffer_utilization"),
         float(i + 1))
     self._assertSummaryContains(
         summary_str,
         self.regexForNodeName("PrefetchDataset", "buffer_capacity"))
     self._assertSummaryContains(
         summary_str, self.regexForNodeName("PrefetchDataset", "buffer_size"))
     self._assertSummaryHasRange(
         summary_str,
         self.regexForNodeName("PrefetchDataset", "buffer_utilization"), 0, 1)
   with self.assertRaises(errors.OutOfRangeError):
     self.evaluate(next_element())
   summary_str = self.evaluate(aggregator.get_summary())
   self._assertSummaryHasCount(
       summary_str,
       self.regexForNodeName("PrefetchDataset", "buffer_utilization"), 100)