Python tensorflow.multinomial方法代码示例

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow 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.square(x - tf.stop_gradient(x_means)))
  return x_means_hot, e_loss 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import multinomial [as 别名]
def multinomial_sample(x, vocab_size, sampling_method, temperature):
  """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 [...].
  """
  if sampling_method == "random":
    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: import tensorflow [as 别名]
# 或者: from tensorflow import multinomial [as 别名]
def vq_nearest_neighbor(x, means, soft_em=False, num_samples=10):
  """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:
    x_means_idx = tf.argmax(-dist, axis=-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.square(x - tf.stop_gradient(x_means)))
  return x_means_hot, e_loss 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import multinomial [as 别名]
def _head(self, policy_input, heading, xy, target_xy):
    """Build the head of the agent: linear policy and value function, and pass
    the auxiliary outputs through.
    """

    # Linear policy and value function.
    policy_logits = snt.Linear(
        self._num_actions, name='policy_logits')(policy_input)
    baseline = tf.squeeze(snt.Linear(1, name='baseline')(policy_input), axis=-1)

    # Sample an action from the policy.
    new_action = tf.multinomial(
        policy_logits, num_samples=1, output_dtype=tf.int32)
    new_action = tf.squeeze(new_action, 1, name='new_action')

    return AgentOutput(
        new_action, policy_logits, baseline, heading, xy, target_xy) 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow 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: import tensorflow [as 别名]
# 或者: from tensorflow 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: import tensorflow [as 别名]
# 或者: from tensorflow import multinomial [as 别名]
def sample_with_temperature(logits, temperature):
  """Either argmax or random sampling.

  Args:
    logits: a Tensor.
    temperature: a float  0.0=argmax 1.0=random

  Returns:
    a Tensor with one fewer dimension than logits.
  """
  if temperature == 0.0:
    # TF argmax doesn't handle >5 dimensions, so we reshape here.
    logits_shape = shape_list(logits)
    argmax = tf.argmax(tf.reshape(logits, [-1, logits_shape[-1]]), axis=1)
    return tf.reshape(argmax, logits_shape[:-1])
  else:
    assert temperature > 0.0
    reshaped_logits = (
        tf.reshape(logits, [-1, shape_list(logits)[-1]]) / temperature)
    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: import tensorflow [as 别名]
# 或者: from tensorflow import multinomial [as 别名]
def sampleAction(self, states):
    # TODO: use this code piece when tf.multinomial gets better
    # sample action from current policy
    # actions = self.session.run(self.predicted_actions, {self.states: states})[0]
    # return actions[0]

    # temporary workaround
    def softmax(y):
      """ simple helper function here that takes unnormalized logprobs """
      maxy = np.amax(y)
      e = np.exp(y - maxy)
      return e / np.sum(e)

    # epsilon-greedy exploration strategy
    if random.random() < self.exploration:
      return random.randint(0, self.num_actions-1)
    else:
      action_scores = self.session.run(self.action_scores, {self.states: states})[0]
      action_probs  = softmax(action_scores) - 1e-5
      action = np.argmax(np.random.multinomial(1, action_probs))
      return action 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import multinomial [as 别名]
def create_softmax_from_logits(logits):
  "Create nodes for softmax computation from logits."
  temperature = tf.placeholder_with_default(
    1.0, shape=(), name='temperature')
  logits = logits / temperature

  logits_shape = tf.shape(logits)
  logits_dim = logits_shape[-1]
  logits_2d = tf.reshape(logits, [-1, logits_dim])
  samples = tf.multinomial(logits_2d, 1)
  samples = tf.reshape(samples, logits_shape[:-1])

  probs = tf.nn.softmax(logits)
  predictions = tf.argmax(probs, axis=2)
    
  return logits, probs, predictions, samples, temperature


# Embedding 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import multinomial [as 别名]
def predict_from_model(logit_groups_geometry, logit_groups_semantics,
                       temperature):
  """Reconstruct predicted geometry and semantics from model output."""
  predictions_geometry_list = []
  for logit_group in logit_groups_geometry:
    if FLAGS.p_norm > 0:
      predictions_geometry_list.append(logit_group[:, :, :, :, 0])
    else:
      logit_group_shape = logit_group.shape_as_list()
      logit_group = tf.reshape(logit_group, [-1, logit_group_shape[-1]])
      samples = tf.multinomial(temperature * logit_group, 1)
      predictions_geometry_list.append(
          tf.reshape(samples, logit_group_shape[:-1]))
  predictions_semantics_list = []
  if FLAGS.predict_semantics:
    for logit_group in logit_groups_semantics:
      predictions_semantics_list.append(tf.argmax(logit_group, 4))
  else:
    predictions_semantics_list = [
        tf.zeros(shape=predictions_geometry_list[0].shape, dtype=tf.uint8)
    ] * len(predictions_geometry_list)
  return predictions_geometry_list, predictions_semantics_list 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import multinomial [as 别名]
def build_forward(self, _input):
		output = _input  # [batch_size, num_steps, rnn_units]
		feature_dim = int(output.get_shape()[2])  # rnn_units
		output = tf.reshape(output, [-1, feature_dim])  # [batch_size * num_steps, rnn_units]
		final_activation = 'sigmoid' if self.out_dim == 1 else 'softmax'
		if self.net_type == 'simple':
			net_config = [] if self.net_config is None else self.net_config
			with tf.variable_scope('wider_actor'):
				for layer in net_config:
					units, activation = layer.get('units'), layer.get('activation', 'relu')
					output = BasicModel.fc_layer(output, units, use_bias=True)
					output = BasicModel.activation(output, activation)
				logits = BasicModel.fc_layer(output, self.out_dim, use_bias=True)  # [batch_size * num_steps, out_dim]
			probs = BasicModel.activation(logits, final_activation)  # [batch_size * num_steps, out_dim]
			probs_dim = self.out_dim
			if self.out_dim == 1:
				probs = tf.concat([1 - probs, probs], axis=1)
				probs_dim = 2
				
			self.decision = tf.multinomial(tf.log(probs), 1)  # [batch_size * num_steps, 1]
			self.decision = tf.reshape(self.decision, [-1, self.num_steps])  # [batch_size, num_steps]
			self.probs = tf.reshape(probs, [-1, self.num_steps, probs_dim])  # [batch_size, num_steps, out_dim]
		else:
			raise ValueError('Do not support %s' % self.net_type) 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow 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)
        reshaped_logits = (
            tf.reshape(logits, [-1, logits_shape[-1]]) / temperature)
        choices = tf.multinomial(reshaped_logits, 1)
        choices = tf.reshape(choices, logits_shape[:-1])
        return choices

      samples = multinomial_squeeze(logits, self.hparams.sampling_temp)

    return samples, logits, losses 

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

  Args:
    latents_pred: Tensor of shape [..., depth].
    latents_discrete_hot: Tensor of shape [..., vocab_size].
    hparams: tf.contrib.training.HParams.

  Returns:
    sample: Tensor of shape [...], a sample from a multinomial distribution.
    loss: Tensor of shape [...], the softmax cross-entropy.
  """
  vocab_size = 2**hparams.bottleneck_bits
  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)
    sample = multinomial_sample(latents_logits,
                                vocab_size,
                                hparams.sampling_method,
                                hparams.sampling_temp)
    return sample, loss 

# 需要导入模块: import tensorflow [as 别名]
# 或者: from tensorflow import multinomial [as 别名]
def _head(self, core_output):
    """Build the head of the agent: linear policy and value function."""
    policy_logits = snt.Linear(
        self._num_actions, name='policy_logits')(
            core_output)
    baseline = tf.squeeze(snt.Linear(1, name='baseline')(core_output), axis=-1)

    # Sample an action from the policy.
    new_action = tf.multinomial(
        policy_logits, num_samples=1, output_dtype=tf.int32)
    new_action = tf.squeeze(new_action, 1, name='new_action')

    return AgentOutput(new_action, policy_logits, baseline)