Python v1.concat方法代码示例

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def collective_group_key(devices):
  """Returns a group key for the set of devices.

  Args:
    devices: list of strings naming devices in a collective group.

  Returns:
    int key uniquely identifying the set of device names.
  """
  global _group_key
  global _group_key_table
  parsed = [pydev.DeviceSpec.from_string(d) for d in devices]
  names = sorted(['%s:%d' % (d.device_type, d.device_index) for d in parsed])
  concat = ','.join(names)
  if concat not in _group_key_table.keys():
    new_key = _group_key
    _group_key += 1
    _group_key_table[concat] = new_key
  rv = _group_key_table[concat]
  return rv 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def _build_tiled_linear(self, inputs, input_name_and_sizes,
                          output_name_and_sizes, add_bias):
    # pylint: disable=missing-docstring
    def split_output(output):
      if len(output_name_and_sizes) == 1:
        return output
      elif len(set([size for _, size in output_name_and_sizes])) == 1:
        # This is a bit faster than several tf.slice calls.
        return tf.split(output, len(output_name_and_sizes), axis=1)
      else:
        outputs = []
        offset = 0
        for _, output_size in output_name_and_sizes:
          outputs.append(tf.slice(output, [0, offset], [-1, output_size]))
          offset += output_size
        return outputs

    weights = self._ensure_weights()
    if len(inputs) > 1:
      inputs = tf.concat(inputs, 1)
    if add_bias:
      biases = self._ensure_biases()
      return split_output(tf.nn.xw_plus_b(inputs, weights, biases))
    else:
      return split_output(tf.matmul(inputs, weights)) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def simulate(self, action):
    reward, done = self._batch_env.simulate(action)
    with tf.control_dependencies([reward, done]):
      new_observ = tf.expand_dims(self._batch_env.observ, axis=1)

      # If we shouldn't stack, i.e. self.history == 1, then just assign
      # new_observ to self._observ and return from here.
      if self.history == 1:
        with tf.control_dependencies([self._observ.assign(new_observ)]):
          return tf.identity(reward), tf.identity(done)

      # If we should stack, then do the required work.
      old_observ = tf.gather(
          self._observ.read_value(),
          list(range(1, self.history)),
          axis=1)
      with tf.control_dependencies([new_observ, old_observ]):
        with tf.control_dependencies([self._observ.assign(
            tf.concat([old_observ, new_observ], axis=1))]):
          return tf.identity(reward), tf.identity(done) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def restore(self, x):
    """Add padding back to the given tensor.

    Args:
      x (tf.Tensor): of shape [dim_compressed,...]

    Returns:
      a tensor of shape [dim_origin,...] with dim_compressed >= dim_origin. The
      dim is restored from the original reference tensor
    """
    with tf.name_scope("pad_reduce/restore"):
      x = tf.scatter_nd(
          indices=self.nonpad_ids,
          updates=x,
          shape=tf.concat([self.dim_origin, tf.shape(x)[1:]], axis=0),
      )
    return x 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def combine(self, expert_out, multiply_by_gates=True):
    """Sum together the expert output, weighted by the gates.

    The slice corresponding to a particular batch element `b` is computed
    as the sum over all experts `i` of the expert output, weighted by the
    corresponding gate values.  If `multiply_by_gates` is set to False, the
    gate values are ignored.

    Args:
      expert_out: a list of `num_experts` `Tensor`s, each with shape
        `[expert_batch_size_i, <extra_output_dims>]`.
      multiply_by_gates: a boolean

    Returns:
      a `Tensor` with shape `[batch_size, <extra_output_dims>]`.
    """
    # see comments on convert_gradient_to_tensor
    stitched = common_layers.convert_gradient_to_tensor(
        tf.concat(expert_out, 0))
    if multiply_by_gates:
      stitched *= tf.expand_dims(self._nonzero_gates, 1)
    combined = tf.unsorted_segment_sum(stitched, self._batch_index,
                                       tf.shape(self._gates)[0])
    return combined 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def call(self, inputs, **kwargs):
    del kwargs
    features = inputs
    set_custom_getter_compose(self._custom_getter)
    tf.get_variable_scope().set_initializer(
        optimize.get_variable_initializer(self.hparams))
    with self._eager_var_store.as_default():
      self._fill_problem_hparams_features(features)
      summarize_features(features, num_shards=self._num_datashards)
      sharded_features = self._shard_features(features)
      sharded_logits, losses = self.model_fn_sharded(sharded_features)
      if isinstance(sharded_logits, dict):
        concat_logits = {}
        for k, v in six.iteritems(sharded_logits):
          concat_logits[k] = tf.concat(v, 0)
        return concat_logits, losses
      else:
        return tf.concat(sharded_logits, 0), losses 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def initialize_write_strengths(self, batch_size):
    """Initialize write strengths which write in both directions.

    Unlike in Grefenstette et al., It's writing out from the center of the
    memory so that it doesn't need to shift the entire memory forward at each
    step.

    Args:
      batch_size: The size of the current batch.

    Returns:
      A tf.float32 tensor of shape [num_write_heads, memory_size, 1].
    """
    memory_center = self._memory_size // 2
    return tf.expand_dims(
        tf.concat([
            # The write strength for the deque bottom.
            # Should be shifted back at each timestep.
            tf.one_hot([[memory_center - 1]] * batch_size,
                       depth=self._memory_size, dtype=tf.float32),
            # The write strength for the deque top.
            # Should be shifted forward at each timestep.
            tf.one_hot([[memory_center]] * batch_size,
                       depth=self._memory_size, dtype=tf.float32)
        ], axis=1), axis=3) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def ae_latent_sample(latents_dense, inputs, ed, embed, iters, hparams):
  """Sample from the latent space in the autoencoder."""
  if hparams.num_decode_blocks < 2 and hparams.sampling_temp == 0.0:
    # TODO(lukaszkaiser): beam-search only works in non-blocked mode for now.
    tf.logging.info("Running beam-search for latents with beam size 1.")
    return ae_latent_sample_beam(latents_dense, inputs, ed, embed, hparams)
  latents_pred = decode_transformer(inputs, ed, latents_dense, hparams, "extra")
  latents_discrete, _ = ae_latent_softmax(latents_pred, None, hparams)

  def next_bit(latents_discrete, i):
    latents_discrete_prev = latents_discrete
    with tf.variable_scope(tf.get_variable_scope(), reuse=True):
      latents_dense = embed(latents_discrete)
      latents_pred = decode_transformer(
          inputs, ed, latents_dense, hparams, "extra")
      latents_discrete, _ = ae_latent_softmax(latents_pred, None, hparams)
      return tf.concat([latents_discrete_prev[:, :(i+1), :],
                        latents_discrete[:, (i+1):, :]], axis=1)

  for i in range(iters):
    latents_discrete = next_bit(latents_discrete, i)
  return latents_discrete 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def add_edge_bias(x, filter_size):
  """Pad x and concatenates an edge bias across the depth of x.

  The edge bias can be thought of as a binary feature which is unity when
  the filter is being convolved over an edge and zero otherwise.

  Args:
    x: Input tensor, shape (NHWC)
    filter_size: filter_size to determine padding.
  Returns:
    x_pad: Input tensor, shape (NHW(c+1))
  """
  x_shape = common_layers.shape_list(x)
  if filter_size[0] == 1 and filter_size[1] == 1:
    return x
  a = (filter_size[0] - 1) // 2  # vertical padding size
  b = (filter_size[1] - 1) // 2  # horizontal padding size
  padding = [[0, 0], [a, a], [b, b], [0, 0]]
  x_bias = tf.zeros(x_shape[:-1] + [1])

  x = tf.pad(x, padding)
  x_pad = tf.pad(x_bias, padding, constant_values=1)
  return tf.concat([x, x_pad], axis=3) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def sample(self, features=None, shape=None):
    del features
    hp = self.hparams
    div_x = 2**hp.num_hidden_layers
    div_y = 1 if self.is1d else 2**hp.num_hidden_layers
    size = [
        hp.batch_size, hp.sample_height // div_x, hp.sample_width // div_y,
        hp.bottleneck_bits
    ]
    size = size if shape is None else shape
    rand = tf.random_uniform(size)
    res = 2.0 * tf.to_float(tf.less(0.5, rand)) - 1.0
    # If you want to set some first bits to a fixed value, do this:
    # fixed = tf.zeros_like(rand) - 1.0
    # nbits = 3
    # res = tf.concat([fixed[:, :, :, :nbits], res[:, :, :, nbits:]], axis=-1)
    return res 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def shake_shake_skip_connection(x, output_filters, stride, is_training):
  """Adds a residual connection to the filter x for the shake-shake model."""
  curr_filters = common_layers.shape_list(x)[-1]
  if curr_filters == output_filters:
    return x
  stride_spec = [1, stride, stride, 1]
  # Skip path 1.
  path1 = tf.nn.avg_pool(x, [1, 1, 1, 1], stride_spec, "VALID")
  path1 = tf.layers.conv2d(
      path1, int(output_filters / 2), (1, 1), padding="SAME", name="path1_conv")

  # Skip path 2.
  pad_arr = [[0, 0], [0, 1], [0, 1], [0, 0]]  # First pad with 0's then crop.
  path2 = tf.pad(x, pad_arr)[:, 1:, 1:, :]
  path2 = tf.nn.avg_pool(path2, [1, 1, 1, 1], stride_spec, "VALID")
  path2 = tf.layers.conv2d(
      path2, int(output_filters / 2), (1, 1), padding="SAME", name="path2_conv")

  # Concat and apply BN.
  final_path = tf.concat(values=[path1, path2], axis=-1)
  final_path = tf.layers.batch_normalization(
      final_path, training=is_training, name="final_path_bn")
  return final_path 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def final_block(x1, x2, dim='2d', training=True, scope='final_block'):
  """Converts activations from last RevNet block to pre-logits.

  Args:
    x1: [NxHxWxC] tensor of network activations.
    x2: [NxHxWxC] tensor of network activations.
    dim: '2d' if 2-dimensional, '3d' if 3-dimensional.
    training: True for train phase, False for eval phase.
    scope: Optional variable scope for the final block.

  Returns:
    [N, hidden_dim] pre-logits tensor from activations x1 and x2.
  """

  # Final batch norm and relu
  with tf.variable_scope(scope):
    y = tf.concat([x1, x2], axis=CONFIG[dim]['split_axis'])
    y = tf.layers.batch_normalization(y, training=training)
    y = tf.nn.relu(y)

    # Global average pooling
    net = tf.reduce_mean(y, CONFIG[dim]['reduction_dimensions'],
                         name='final_pool', keep_dims=True)

    return net 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def inject_latent(self, layer, inputs, target, action):
    """Inject a VAE-style latent."""
    del action
    # Latent for stochastic model
    filters = 128
    full_video = tf.stack(inputs + [target], axis=1)
    latent_mean, latent_std = self.construct_latent_tower(
        full_video, time_axis=1)
    latent = common_video.get_gaussian_tensor(latent_mean, latent_std)
    latent = tfl.flatten(latent)
    latent = tf.expand_dims(latent, axis=1)
    latent = tf.expand_dims(latent, axis=1)
    latent_mask = tfl.dense(latent, filters, name="latent_mask")
    zeros_mask = tf.zeros(
        common_layers.shape_list(layer)[:-1] + [filters], dtype=tf.float32)
    layer = tf.concat([layer, latent_mask + zeros_mask], axis=-1)
    extra_loss = self.get_kl_loss([latent_mean], [latent_std])
    return layer, extra_loss 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def update_internal_states_early(self, internal_states, frames):
    """Update the internal states early in the network in GRU-like way."""
    batch_size = common_layers.shape_list(frames[0])[0]
    internal_state = internal_states[0][0][:batch_size, :, :, :]
    state_activation = tf.concat([internal_state, frames[0]], axis=-1)
    state_gate_candidate = tf.layers.conv2d(
        state_activation, 2 * self.hparams.recurrent_state_size,
        (3, 3), padding="SAME", name="state_conv")
    state_gate, state_candidate = tf.split(state_gate_candidate, 2, axis=-1)
    state_gate = tf.nn.sigmoid(state_gate)
    state_candidate = tf.tanh(state_candidate)
    internal_state = internal_state * state_gate
    internal_state += state_candidate * (1.0 - state_gate)
    max_batch_size = max(_MAX_BATCH, self.hparams.batch_size)
    diff_batch_size = max_batch_size - batch_size
    internal_state = tf.pad(
        internal_state, [[0, diff_batch_size], [0, 0], [0, 0], [0, 0]])
    return [[internal_state]] 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import concat [as 别名]
def simple_discrete_latent_tower(self, input_image, target_image):
    hparams = self.hparams

    if self.is_predicting:
      batch_size = common_layers.shape_list(input_image)[0]
      rand = tf.random_uniform([batch_size, hparams.bottleneck_bits])
      bits = 2.0 * tf.to_float(tf.less(0.5, rand)) - 1.0
      return bits

    conv_size = self.tinyify([64, 32, 32, 1])
    pair = tf.concat([input_image, target_image], axis=-1)
    posterior_enc = self.basic_conv_net(pair, conv_size, "posterior_enc")
    posterior_enc = tfl.flatten(posterior_enc)
    bits, _ = discretization.tanh_discrete_bottleneck(
        posterior_enc,
        hparams.bottleneck_bits,
        hparams.bottleneck_noise,
        hparams.discretize_warmup_steps,
        hparams.mode)
    return bits