Python v1.multinomial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 multinomial [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 multinomial [as 别名]
def sample_temperature_per_example(logits, temperature, sampling_keep_top_k=-1):
  """Either random sampling with different temperature per example.

  Args:
    logits: a Tensor.
    temperature: a float vector of same size as logits.
    sampling_keep_top_k: If not -1, only sample from the top k logits.
  Returns:
    a Tensor with one fewer dimension than logits.
  """
  logits = _select_top_k(logits, sampling_keep_top_k)
  logits /= tf.reshape(temperature, [-1] + [1] * (len(logits.shape) - 1))
  reshaped_logits = tf.reshape(logits, [-1, shape_list(logits)[-1]])
  choices = tf.multinomial(reshaped_logits, 1)
  choices = tf.reshape(choices,
                       shape_list(logits)[:logits.get_shape().ndims - 1])
  return choices 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def multinomial_sample(x, vocab_size=None, sampling_method="random",
                       temperature=1.0):
  """Multinomial sampling from a n-dimensional tensor.

  Args:
    x: Tensor of shape [..., vocab_size]. Parameterizes logits of multinomial.
    vocab_size: Number of classes in multinomial distribution.
    sampling_method: String, "random" or otherwise deterministic.
    temperature: Positive float.

  Returns:
    Tensor of shape [...].
  """
  vocab_size = vocab_size or common_layers.shape_list(x)[-1]
  if sampling_method == "random" and temperature > 0.0:
    samples = tf.multinomial(tf.reshape(x, [-1, vocab_size]) / temperature, 1)
  else:
    samples = tf.argmax(x, axis=-1)
  reshaped_samples = tf.reshape(samples, common_layers.shape_list(x)[:-1])
  return reshaped_samples 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def categorical_sample(logits, dtype=tf.int32,
                       sample_shape=(), seed=None):
  """Samples from categorical distribution."""
  logits = tf.convert_to_tensor(logits, name="logits")
  event_size = tf.shape(logits)[-1]
  batch_shape_tensor = tf.shape(logits)[:-1]
  def _sample_n(n):
    """Sample vector of categoricals."""
    if logits.shape.ndims == 2:
      logits_2d = logits
    else:
      logits_2d = tf.reshape(logits, [-1, event_size])
    sample_dtype = tf.int64 if logits.dtype.size > 4 else tf.int32
    draws = tf.multinomial(
        logits_2d, n, seed=seed, output_dtype=sample_dtype)
    draws = tf.reshape(
        tf.transpose(draws),
        tf.concat([[n], batch_shape_tensor], 0))
    return tf.cast(draws, dtype)
  return _call_sampler(_sample_n, sample_shape) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def sample(self, features):
    """Run the model and extract samples.

    Args:
      features: an map of string to `Tensor`.

    Returns:
       samples: an integer `Tensor`.
       logits: a list of `Tensor`s, one per datashard.
       losses: a dictionary: {loss-name (string): floating point `Scalar`}.
    """
    logits, losses = self(features)  # pylint: disable=not-callable
    if self._target_modality_is_real:
      return logits, logits, losses  # Raw numbers returned from real modality.
    if self.hparams.sampling_method == "argmax":
      samples = tf.argmax(logits, axis=-1)
    else:
      assert self.hparams.sampling_method == "random"

      def multinomial_squeeze(logits, temperature=1.0):
        logits_shape = common_layers.shape_list(logits)
        logits /= tf.reshape(temperature, [-1] + [1] * (len(logits_shape) - 1))
        reshaped_logits = tf.reshape(logits, [-1, logits_shape[-1]])
        choices = tf.multinomial(reshaped_logits, 1)
        choices = tf.reshape(choices, logits_shape[:-1])
        return choices

      temperature = features.get("sampling_temp", self.hparams.sampling_temp)
      samples = multinomial_squeeze(logits, temperature)

    return samples, logits, losses 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def multinomial_sample(x, vocab_size, temperature):
  """Multinomial sampling from a n-dimensional tensor."""
  if temperature > 0:
    samples = tf.multinomial(tf.reshape(x, [-1, vocab_size]) / temperature, 1)
  else:
    samples = tf.argmax(x, axis=-1)
  reshaped_samples = tf.reshape(samples, common_layers.shape_list(x)[:-1])
  return tf.to_int32(reshaped_samples) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def video_pixel_noise_bottom(x, model_hparams, vocab_size):
  """Bottom transformation for video."""
  input_noise = getattr(model_hparams, "video_modality_input_noise", 0.25)
  inputs = x
  if model_hparams.mode == tf.estimator.ModeKeys.TRAIN:
    background = tfp.stats.percentile(inputs, 50., axis=[0, 1, 2, 3])
    input_shape = common_layers.shape_list(inputs)
    input_size = tf.reduce_prod(input_shape[:-1])
    input_mask = tf.multinomial(
        tf.log([[input_noise, 1.-input_noise]]), input_size)
    input_mask = tf.reshape(tf.cast(input_mask, tf.int32),
                            input_shape[:-1]+[1])
    inputs = inputs * input_mask + background * (1 - input_mask)
  return video_bottom(inputs, model_hparams, vocab_size) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def ae_latent_softmax(latents_pred, latents_discrete_hot, vocab_size, hparams):
  """Latent prediction and loss.

  Args:
    latents_pred: Tensor of shape [..., depth].
    latents_discrete_hot: Tensor of shape [..., vocab_size].
    vocab_size: an int representing the vocab size.
    hparams: HParams.

  Returns:
    sample: Tensor of shape [...], a sample from a multinomial distribution.
    loss: Tensor of shape [...], the softmax cross-entropy.
  """
  with tf.variable_scope("latent_logits"):
    latents_logits = tf.layers.dense(latents_pred, vocab_size,
                                     name="logits_dense")
    if hparams.logit_normalization:
      latents_logits *= tf.rsqrt(1e-8 +
                                 tf.reduce_mean(tf.square(latents_logits)))
    loss = tf.nn.softmax_cross_entropy_with_logits_v2(
        labels=latents_discrete_hot, logits=latents_logits)

    # TODO(trandustin): tease this out from ae_latent_softmax.
    # we use just the loss portion to anchor prior / encoder on text.
    sample = multinomial_sample(latents_logits,
                                vocab_size,
                                hparams.sampling_method,
                                hparams.sampling_temp)
    return sample, loss 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def vq_nearest_neighbor(x, means,
                        soft_em=False, num_samples=10, temperature=None):
  """Find the nearest element in means to elements in x."""
  bottleneck_size = common_layers.shape_list(means)[0]
  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 soft_em:
    x_means_idx = tf.multinomial(-dist, num_samples=num_samples)
    x_means_hot = tf.one_hot(
        x_means_idx, depth=common_layers.shape_list(means)[0])
    x_means_hot = tf.reduce_mean(x_means_hot, axis=1)
  else:
    if temperature is None:
      x_means_idx = tf.argmax(-dist, axis=-1)
    else:
      x_means_idx = tf.multinomial(- dist / temperature, 1)
      x_means_idx = tf.squeeze(x_means_idx, axis=-1)
    if (common_layers.should_generate_summaries() and
        not common_layers.is_xla_compiled()):
      tf.summary.histogram("means_idx", tf.reshape(x_means_idx, [-1]))
    x_means_hot = tf.one_hot(x_means_idx, bottleneck_size)
  x_means_hot_flat = tf.reshape(x_means_hot, [-1, bottleneck_size])
  x_means = tf.matmul(x_means_hot_flat, means)
  e_loss = tf.reduce_mean(tf.squared_difference(x, tf.stop_gradient(x_means)))
  return x_means_hot, e_loss, dist 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def provide_one_hot_labels(self, batch_size):
    """Provides one hot labels."""
    pitch_counts = self.get_pitch_counts()
    pitches = sorted(pitch_counts.keys())
    counts = [pitch_counts[p] for p in pitches]
    indices = tf.reshape(
        tf.multinomial(tf.log([tf.to_float(counts)]), batch_size), [batch_size])
    one_hot_labels = tf.one_hot(indices, depth=len(pitches))
    return one_hot_labels 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def sample_with_temperature(logits, temperature):
  """Either argmax after softmax or random sample along the pitch axis.

  Args:
    logits: a Tensor of shape (batch, time, pitch, instrument).
    temperature: a float  0.0=argmax 1.0=random

  Returns:
    a Tensor of the same shape, with one_hots on the pitch dimension.
  """
  logits = tf.transpose(logits, [0, 1, 3, 2])
  pitch_range = tf.shape(logits)[-1]

  def sample_from_logits(logits):
    with tf.control_dependencies([tf.assert_greater(temperature, 0.0)]):
      logits = tf.identity(logits)
    reshaped_logits = (
        tf.reshape(logits, [-1, tf.shape(logits)[-1]]) / temperature)
    choices = tf.multinomial(reshaped_logits, 1)
    choices = tf.reshape(choices,
                         tf.shape(logits)[:logits.get_shape().ndims - 1])
    return choices

  choices = tf.cond(tf.equal(temperature, 0.0),
                    lambda: tf.argmax(tf.nn.softmax(logits), -1),
                    lambda: sample_from_logits(logits))
  samples_onehot = tf.one_hot(choices, pitch_range)
  return tf.transpose(samples_onehot, [0, 1, 3, 2]) 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def ae_latent_softmax(latents_pred, latents_discrete, hparams):
  """Latent prediction and loss."""
  vocab_size = 2 ** hparams.z_size
  if hparams.num_decode_blocks < 2:
    latents_logits = tf.layers.dense(latents_pred, vocab_size,
                                     name="extra_logits")
    if hparams.logit_normalization:
      latents_logits *= tf.rsqrt(1e-8 +
                                 tf.reduce_mean(tf.square(latents_logits)))

    loss = None
    if latents_discrete is not None:
      if hparams.soft_em:
        # latents_discrete is actually one-hot of multinomial samples
        assert hparams.num_decode_blocks == 1
        loss = tf.nn.softmax_cross_entropy_with_logits_v2(
            labels=latents_discrete, logits=latents_logits)
      else:
        loss = tf.nn.sparse_softmax_cross_entropy_with_logits(
            labels=latents_discrete, logits=latents_logits)
    sample = multinomial_sample(
        latents_logits, vocab_size, hparams.sampling_temp)
    return sample, loss

  # Multi-block case.
  vocab_bits = int(math.log(vocab_size, 2))
  assert vocab_size == 2**vocab_bits
  assert vocab_bits % hparams.num_decode_blocks == 0
  block_vocab_size = 2**(vocab_bits // hparams.num_decode_blocks)
  latents_logits = [
      tf.layers.dense(
          latents_pred, block_vocab_size, name="extra_logits_%d" % i)
      for i in range(hparams.num_decode_blocks)
  ]
  loss = None
  if latents_discrete is not None:
    losses = []
    for i in range(hparams.num_decode_blocks):
      d = tf.floormod(tf.floordiv(latents_discrete,
                                  block_vocab_size**i), block_vocab_size)
      losses.append(tf.nn.sparse_softmax_cross_entropy_with_logits(
          labels=d, logits=latents_logits[i]))
    loss = sum(losses)
  samples = [multinomial_sample(l, block_vocab_size, hparams.sampling_temp)
             for l in latents_logits]
  sample = sum([s * block_vocab_size**i for i, s in enumerate(samples)])
  return sample, loss 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def nearest_neighbor(self, x, means):
    """Find the nearest element in means to elements in x.

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

    Returns:
      Tensor with nearest element in mean encoded in one-hot notation.
    """
    x_norm_sq = tf.reduce_sum(tf.square(x), axis=-1, keep_dims=True)
    means_norm_sq = tf.reduce_sum(tf.square(means), axis=-1, keep_dims=True)
    scalar_prod = tf.matmul(
        tf.transpose(x, perm=[1, 0, 2]), tf.transpose(means, perm=[0, 2, 1]))
    scalar_prod = tf.transpose(scalar_prod, perm=[1, 0, 2])
    dist = x_norm_sq + tf.transpose(
        means_norm_sq, perm=[2, 0, 1]) - 2 * scalar_prod

    if self.hparams.soft_em:
      nearest_idx = tf.stack(
          [
              tf.multinomial(
                  -dist[:, i, :], num_samples=self.hparams.num_samples)
              for i in range(self.hparams.num_blocks)
          ],
          axis=1)
      nearest_hot = tf.one_hot(nearest_idx, depth=self.hparams.block_v_size)
      nearest_hot = tf.reduce_mean(nearest_hot, axis=-2)
    else:
      if self.hparams.random_top_k > 1:
        _, top_k_idx = tf.nn.top_k(-dist, k=self.hparams.random_top_k)
        nearest_idx = tf.gather(
            top_k_idx,
            tf.random_uniform(
                [1],
                minval=0,
                maxval=self.hparams.random_top_k - 1,
                dtype=tf.int32),
            axis=-1)
      else:
        if self.hparams.use_scales:
          dist /= tf.reshape(self.hparams.scales,
                             [1, 1, self.hparams.moe_num_experts])
        nearest_idx = tf.argmax(-dist, axis=-1)
      nearest_hot = tf.one_hot(nearest_idx, self.hparams.block_v_size)
    return nearest_hot 

# 需要导入模块: from tensorflow.compat import v1 [as 别名]
# 或者: from tensorflow.compat.v1 import multinomial [as 别名]
def _preprocess(self, features):
    """Preprocesses features for multilingual translation."""
    seqs, tags = {}, {}

    if self._hparams.mode == tf.estimator.ModeKeys.TRAIN:
      seqs["src"] = features["inputs"]
      seqs["tgt"] = features["targets"]
      seqs["aux"] = None
      tags["src"] = features["input_tags"]
      tags["tgt"] = features["target_tags"]
      tags["aux"] = None

      # Construct a tensor of auxiliary tags.
      batch_size = common_layers.shape_list(features["all_tags"])[0]
      num_all_tags = common_layers.shape_list(features["all_tags"])[1]
      # <float32> [num_all_tags, 1, emb_dim].
      all_tags = features["all_tags"][0]  # batch elements are identical.
      # <int32> [batch_size].
      aux_tag_index = tf.multinomial(
          tf.ones([1, num_all_tags]), batch_size,
          output_dtype=tf.int32)[0]
      # <float32> [batch_size, 1, 1, emb_dim].
      tags["aux"] = tf.expand_dims(tf.gather(all_tags, aux_tag_index), 1)

      from_domains = ["src", "src", "tgt"]
      to_domains = ["tgt", "aux", "aux"]
    else:
      seqs["src"] = features["inputs"]
      seqs["tgt"] = features["targets"]
      tags["src"] = None
      tags["tgt"] = features["target_tags"]

      # Expand target tags to beam width, if necessary.
      if self._hparams.mode == tf.estimator.ModeKeys.PREDICT:
        tags["tgt"] = tf.tile(tags["tgt"], [self._hparams.beam_width, 1, 1, 1])

      from_domains = ["src"]
      to_domains = ["tgt"]

    # Construct inputs and targets.
    inputs, targets = {}, {}
    for fd, td in zip(from_domains, to_domains):
      key = "%s>%s" % (fd, td)
      inputs[key], targets[key] = self._build_inputs_and_targets(
          seqs[fd], tags[fd], seqs[td], tags[td])

    return inputs, targets