Python v1.matmul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 matmul [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 matmul [as 别名]
def vq_nearest_neighbor(x, hparams):
  """Find the nearest element in means to elements in x."""
  bottleneck_size = 2**hparams.bottleneck_bits
  means = hparams.means
  x_norm_sq = tf.reduce_sum(tf.square(x), axis=-1, keepdims=True)
  means_norm_sq = tf.reduce_sum(tf.square(means), axis=-1, keepdims=True)
  scalar_prod = tf.matmul(x, means, transpose_b=True)
  dist = x_norm_sq + tf.transpose(means_norm_sq) - 2 * scalar_prod
  if hparams.bottleneck_kind == "em":
    x_means_idx = tf.multinomial(-dist, num_samples=hparams.num_samples)
    x_means_hot = tf.one_hot(
        x_means_idx, depth=bottleneck_size)
    x_means_hot = tf.reduce_mean(x_means_hot, axis=1)
  else:
    x_means_idx = tf.argmax(-dist, axis=-1)
    x_means_hot = tf.one_hot(x_means_idx, depth=bottleneck_size)
  x_means = tf.matmul(x_means_hot, means)
  e_loss = tf.reduce_mean(tf.squared_difference(x, tf.stop_gradient(x_means)))
  return x_means_hot, e_loss 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def attn(image_feat, query, hparams, name="attn"):
  """Attention on image feature with question as query."""
  with tf.variable_scope(name, "attn", values=[image_feat, query]):
    attn_dim = hparams.attn_dim
    num_glimps = hparams.num_glimps
    num_channels = common_layers.shape_list(image_feat)[-1]
    if len(common_layers.shape_list(image_feat)) == 4:
      image_feat = common_layers.flatten4d3d(image_feat)
    query = tf.expand_dims(query, 1)
    image_proj = common_attention.compute_attention_component(
        image_feat, attn_dim, name="image_proj")
    query_proj = common_attention.compute_attention_component(
        query, attn_dim, name="query_proj")
    h = tf.nn.relu(image_proj + query_proj)
    h_proj = common_attention.compute_attention_component(
        h, num_glimps, name="h_proj")
    p = tf.nn.softmax(h_proj, axis=1)
    image_ave = tf.matmul(image_feat, p, transpose_a=True)
    image_ave = tf.reshape(image_ave, [-1, num_channels*num_glimps])

    return image_ave 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def compute_last_embedding(input_embeddings, input_lengths, hparams):
  """Computes average of last K embedding.

  Args:
    input_embeddings: <tf.float32>[bs, max_seq_len, emb_dim]
    input_lengths: <tf.int64>[bs, 1]
    hparams: model hparams

  Returns:
    last_k_embedding: <tf.float32>[bs, emb_dim]
  """
  max_seq_len = tf.shape(input_embeddings)[1]
  # <tf.float32>[bs, 1, max_seq_len]
  mask = tf.sequence_mask(input_lengths, max_seq_len, dtype=tf.float32)
  del_mask = tf.sequence_mask(
      input_lengths - hparams.last_k, max_seq_len, dtype=tf.float32)
  final_mask = mask - del_mask
  # <tf.float32>[bs, 1, emb_dim]
  sum_embedding = tf.matmul(final_mask, input_embeddings)
  # <tf.float32>[bs, 1, emb_dim]
  last_k_embedding = sum_embedding / tf.to_float(
      tf.expand_dims(
          tf.ones([tf.shape(input_embeddings)[0], 1]) * hparams.last_k, 2))
  # <tf.float32>[bs, dim]
  return tf.squeeze(last_k_embedding, 1) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def compute_average_embedding(input_embeddings, input_lengths):
  """Computes bag-of-words embedding.

  Args:
    input_embeddings: <tf.float32>[bs, max_seq_len, emb_dim]
    input_lengths: <tf.int64>[bs, 1]

  Returns:
    bow_embedding: <tf.float32>[bs, emb_dim]
  """
  max_seq_len = tf.shape(input_embeddings)[1]
  # <tf.float32>[bs, 1, max_seq_len]
  mask = tf.sequence_mask(input_lengths, max_seq_len, dtype=tf.float32)
  # <tf.float32>[bs, 1, emb_dim]
  sum_embedding = tf.matmul(mask, input_embeddings)
  # <tf.float32>[bs, 1, emb_dim]
  avg_embedding = sum_embedding / tf.to_float(tf.expand_dims(input_lengths, 2))
  # <tf.float32>[bs, dim]
  return tf.squeeze(avg_embedding, 1) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def smoothing_cross_entropy_factored(a, b, labels, confidence):
  """Memory-efficient computation of smoothing cross-entropy.

  Avoids realizing the entire logits matrix at once.

  Args:
    a: a Tensor with shape [batch, inner_dim]
    b: a Tensor with shape [vocab_size, inner_dim]
    labels: an integer Tensor with shape [batch]
    confidence: a float

  Returns:
    A Tensor with shape [batch]
  """
  num_splits = 16
  vocab_size = shape_list(b)[0]
  labels = approximate_split(labels, num_splits)
  a = approximate_split(a, num_splits)
  parts = []
  for part in range(num_splits):
    with tf.control_dependencies(parts[-1:]):
      logits = tf.matmul(a[part], b, transpose_b=True)
      parts.append(
          smoothing_cross_entropy(logits, labels[part], vocab_size, confidence))
  return tf.concat(parts, 0) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def dense_weightnorm(
    name, x, n_out, x_mask, init_scale, init, dtype=tf.float32):
  """Dense layer with weight normalization."""
  n_in = common_layers.shape_list(x)[2]
  eps = tf.keras.backend.epsilon()
  with tf.variable_scope(name, reuse=tf.AUTO_REUSE):
    v = tf.get_variable(
        "v", [n_in, n_out], dtype,
        initializer=tf.random_normal_initializer(0, 0.05), trainable=True)
    v = v / tf.norm(v, axis=0, keepdims=True)
    t = tf.matmul(x, v)  # [B, L, n_out]
    mean, var = moments_over_bl(t, x_mask)
    g_init = init_scale / (tf.sqrt(var) + eps)
    g = get_variable_ddi(
        "g", [n_out], g_init, init,
        initializer=tf.zeros_initializer, dtype=dtype, trainable=True)
    b = get_variable_ddi(
        "b", [n_out], -mean*g_init, init,
        initializer=tf.zeros_initializer, dtype=dtype, trainable=True)
    w = g * v
    y = tf.matmul(x, w) + b
    tf.summary.histogram("_g", g)
    return y 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def embedding_lookup(self, x, means):
    """Compute nearest neighbors and loss for training the embeddings.

    Args:
        x: Batch of encoder continuous latent states sliced/projected into
        shape
        [-1, num_blocks, block_dim].
        means: Embedding means.

    Returns:
        The nearest neighbor in one hot form, the nearest neighbor
        itself, the
        commitment loss, embedding training loss.
    """
    x_means_hot = self.nearest_neighbor(x, means)
    x_means_hot_flat = tf.reshape(
        x_means_hot, [-1, self.hparams.num_blocks, self.hparams.block_v_size])
    x_means = tf.matmul(tf.transpose(x_means_hot_flat, perm=[1, 0, 2]), means)
    x_means = tf.transpose(x_means, [1, 0, 2])
    q_loss = tf.reduce_mean(
        tf.squared_difference(tf.stop_gradient(x), x_means))
    e_loss = tf.reduce_mean(
        tf.squared_difference(x, tf.stop_gradient(x_means)))
    return x_means_hot, x_means, q_loss, e_loss 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def compute_values(edge_compatibility, v):
  """Compute values. If edge compatibilities is just adjacency, we get ggnn.

  Args:
    edge_compatibility: A tensor of shape [batch, num_transforms, length, depth]
    v: A tensor of shape [batch, num_transforms, length, depth]

  Returns:
    output: A [batch, length, depth] tensor
  """

  # Computes the incoming value vectors for each node by weighting them
  # according to the attention weights. These values are still segregated by
  # edge type.
  # Shape = [B, T, N, V].
  all_edge_values = tf.matmul(tf.to_float(edge_compatibility), v)

  # Combines the weighted value vectors together across edge types into a
  # single N x V matrix for each batch.
  output = tf.reduce_sum(all_edge_values, axis=1)  # Shape [B, N, V].
  return output 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def _address_content(self, x):
    """Address the memory based on content similarity.

    Args:
      x: a tensor in the shape of [batch_size, length, depth].
    Returns:
      the logits for each memory entry [batch_size, length, memory_size].
    """
    mem_keys = tf.layers.dense(self.mem_vals, self.key_depth,
                               bias_initializer=tf.constant_initializer(1.0),
                               name="mem_key")
    mem_query = tf.layers.dense(x, self.key_depth,
                                bias_initializer=tf.constant_initializer(1.0),
                                name="mem_query")
    norm = tf.matmul(self._norm(mem_query), self._norm(mem_keys),
                     transpose_b=True)
    dot_product = tf.matmul(mem_query, mem_keys, transpose_b=True)
    cos_dist = tf.div(dot_product, norm + 1e-7, name="cos_dist")
    access_logits = self.sharpen_factor * cos_dist
    return access_logits 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def testFlopRegularizerDontConvertToVariable(self):
    tf.reset_default_graph()
    tf.set_random_seed(1234)

    x = tf.constant(1.0, shape=[2, 6], name='x', dtype=tf.float32)
    w = tf.Variable(tf.truncated_normal([6, 4], stddev=1.0), use_resource=True)
    net = tf.matmul(x, w)

    # Create FLOPs network regularizer.
    threshold = 0.9
    flop_reg = flop_regularizer.GroupLassoFlopsRegularizer([net.op], threshold,
                                                           0)

    with self.cached_session():
      tf.global_variables_initializer().run()
      flop_reg.get_regularization_term().eval() 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def testMatMul2D(self, size):
    inputs = tf.zeros((13, 2))
    handler = matmul_source_op_handler.MatMulSourceOpHandler(0.1)

    kernel = tf.constant([[1, 2, 3], [4, 5, 6]], dtype=tf.float32)
    x = tf.matmul(inputs, kernel, transpose_b=False, name='MatMul')
    op_slice = orm.OpSlice(x.op, orm.Slice(0, size))

    transpose_kernel = tf.constant([[1, 4], [2, 5], [3, 6]], dtype=tf.float32)
    x_other = tf.matmul(
        inputs,
        transpose_kernel,
        transpose_b=True,
        name='MatMulTransposedKernel')
    op_slice_other = orm.OpSlice(x_other.op, orm.Slice(0, size))

    self.assertAllClose(
        handler.create_regularizer(op_slice).regularization_vector,
        handler.create_regularizer(op_slice_other).regularization_vector) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def get_mlm_logits(input_tensor, albert_config, mlm_positions, output_weights):
  """From run_pretraining.py."""
  input_tensor = gather_indexes(input_tensor, mlm_positions)
  with tf.variable_scope("cls/predictions"):
    # We apply one more non-linear transformation before the output layer.
    # This matrix is not used after pre-training.
    with tf.variable_scope("transform"):
      input_tensor = tf.layers.dense(
          input_tensor,
          units=albert_config.embedding_size,
          activation=modeling.get_activation(albert_config.hidden_act),
          kernel_initializer=modeling.create_initializer(
              albert_config.initializer_range))
      input_tensor = modeling.layer_norm(input_tensor)

    # The output weights are the same as the input embeddings, but there is
    # an output-only bias for each token.
    output_bias = tf.get_variable(
        "output_bias",
        shape=[albert_config.vocab_size],
        initializer=tf.zeros_initializer())
    logits = tf.matmul(
        input_tensor, output_weights, transpose_b=True)
    logits = tf.nn.bias_add(logits, output_bias)
  return logits 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def get_sentence_order_logits(input_tensor, albert_config):
  """Get loss and log probs for the next sentence prediction."""

  # Simple binary classification. Note that 0 is "next sentence" and 1 is
  # "random sentence". This weight matrix is not used after pre-training.
  with tf.variable_scope("cls/seq_relationship"):
    output_weights = tf.get_variable(
        "output_weights",
        shape=[2, albert_config.hidden_size],
        initializer=modeling.create_initializer(
            albert_config.initializer_range))
    output_bias = tf.get_variable(
        "output_bias", shape=[2], initializer=tf.zeros_initializer())

    logits = tf.matmul(input_tensor, output_weights, transpose_b=True)
    logits = tf.nn.bias_add(logits, output_bias)
    return logits 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import matmul [as 别名]
def get_mlm_logits(model, albert_config, mlm_positions):
  """From run_pretraining.py."""
  input_tensor = gather_indexes(model.get_sequence_output(), mlm_positions)
  with tf.variable_scope("cls/predictions"):
    # We apply one more non-linear transformation before the output layer.
    # This matrix is not used after pre-training.
    with tf.variable_scope("transform"):
      input_tensor = tf.layers.dense(
          input_tensor,
          units=albert_config.embedding_size,
          activation=modeling.get_activation(albert_config.hidden_act),
          kernel_initializer=modeling.create_initializer(
              albert_config.initializer_range))
      input_tensor = modeling.layer_norm(input_tensor)

    # The output weights are the same as the input embeddings, but there is
    # an output-only bias for each token.
    output_bias = tf.get_variable(
        "output_bias",
        shape=[albert_config.vocab_size],
        initializer=tf.zeros_initializer())
    logits = tf.matmul(
        input_tensor, model.get_embedding_table(), transpose_b=True)
    logits = tf.nn.bias_add(logits, output_bias)
  return logits