Python tensorflow.DType方法代碼示例

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def getEstimator(cls, priors: Tuple[Distribution, ...], K: int,
                     dtype: tf.DType = tf.float32,
                     isFullyObserved: bool = True,
                     noiseUniformity: NoiseUniformity = HOMOGENEOUS,
                     stopCriterionInit=LlhStall(10),
                     stopCriterionEM=LlhStall(100),
                     stopCriterionBCD=LlhImprovementThreshold(1e-2),
                     path: str = "/tmp", device: str = "/cpu:0",
                     cv: CV = None):

        def model_fn(features, labels, mode):
            es = cls.__estimatorSpec(mode=mode, features=features,
                                     isFullyObserved=isFullyObserved,
                                     device=device, priors=priors,
                                     noiseUniformity=noiseUniformity,
                                     stopCriterionInit=stopCriterionInit,
                                     stopCriterionEM=stopCriterionEM,
                                     stopCriterionBCD=stopCriterionBCD,
                                     cv=cv, path=path, K=K,
                                     transform=False, dtype=dtype)
            return(es)

        est = tf.estimator.Estimator(model_fn=model_fn,
                                     model_dir=path)
        return(est) 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def parameterInfo(self,
                      shape: Tuple[int, ...] = (1,),
                      latentShape: Tuple[int, ...] = ()) -> ParameterInfo:
        """Initializers of the parameters of the distribution.

        Draw random initialization values for each parameter matching the
        provided `shape`, `lantentShape`, and `dtype`. This method has to
        be implemented by concrete distributions to provide reasonable
        random initalizations used during `Distribution.random`.

        Arguments:
            shape: `Tuple[int, ...]` the shape of the distribution.
            latentShape: `Tuple[int, ...]` the latent shape of the
                distribution.
            dtype: `DType` the data type of the distribution.

        Returns:
            `Dict[str, Tensor]` map from parameter names to `Tensor` s
            containing the random initial values for the parameters.
        """
        ... 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def __init__(self, in_layers=None, dtype=None, **kwargs):
    """
    Parameters
    ----------
    dtype: tf.DType
      the dtype to cast the in_layer to
      e.x. tf.int32
    """
    if dtype is None:
      raise ValueError("Must cast to a dtype")
    self.dtype = dtype
    super(Cast, self).__init__(in_layers, **kwargs)
    try:
      parent_shape = self.in_layers[0].shape
      self._shape = parent_shape
    except:
      pass 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def decode_field(self,
                     field_name: str,
                     field_value: Union[tf.Tensor, tf.SparseTensor],
                     field_type: Optional[tf.DType] = None) -> tf.Tensor:
        """
        Decode a field from a tfrecord example

        Parameters
        ----------
        field_name
            name of the field, if nested - will be separated using "/"
        field_value
            value of the field from tfrecords example
        field_type
            type of the decoded field from self.get_tfrecords_output_types
            or None, if it was not provided
        """
        # pylint: disable=no-self-use
        # is designed to be overridden
        # pylint: disable=unused-argument
        # this method is really an interface, but has a default implementation.
        if field_type is None:
            return field_value
        return tf.decode_raw(field_value, field_type) 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def layer_norm(inputs, epsilon=1e-6, dtype=None, scope=None):
    """
    Layer Normalization
    :param inputs: A Tensor of shape [..., channel_size]
    :param epsilon: A floating number
    :param dtype: An optional instance of tf.DType
    :param scope: An optional string
    :returns: A Tensor with the same shape as inputs
    """
    with tf.variable_scope(scope, default_name="layer_norm", values=[inputs],
                           dtype=dtype):
        channel_size = inputs.get_shape().as_list()[-1]

        scale = tf.get_variable("scale", shape=[channel_size],
                                initializer=tf.ones_initializer())

        offset = tf.get_variable("offset", shape=[channel_size],
                                 initializer=tf.zeros_initializer())

        mean = tf.reduce_mean(inputs, -1, True)
        variance = tf.reduce_mean(tf.square(inputs - mean), -1, True)

        norm_inputs = (inputs - mean) * tf.rsqrt(variance + epsilon)

        return norm_inputs * scale + offset 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def select_eps_for_addition(dtype):
  """Returns 2 * machine epsilon based on `dtype`.

  This function picks an epsilon slightly greater than the machine epsilon,
  which is the upper bound on relative error. This value ensures that
  `1.0 + eps != 1.0`.

  Args:
    dtype: The `tf.DType` of the tensor to which eps will be added.

  Raises:
    ValueError: If `dtype` is not a floating type.

  Returns:
    A `float` to be used to make operations safe.
  """
  return 2.0 * np.finfo(dtype.as_numpy_dtype).eps 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def select_eps_for_division(dtype):
  """Selects default values for epsilon to make divisions safe based on dtype.

  This function returns an epsilon slightly greater than the smallest positive
  floating number that is representable for the given dtype. This is mainly used
  to prevent division by zero, which produces Inf values. However, if the
  nominator is orders of magnitude greater than `1.0`, eps should also be
  increased accordingly. Only floating types are supported.

  Args:
    dtype: The `tf.DType` of the tensor to which eps will be added.

  Raises:
    ValueError: If `dtype` is not a floating type.

  Returns:
    A `float` to be used to make operations safe.
  """
  return 10.0 * np.finfo(dtype.as_numpy_dtype).tiny


# The util functions or classes are not exported. 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def __init__(self, shape, dtype=np.float32):
    """Creates a `ShapeDtype` instance, with canonicalized `shape` and `dtype`.

    Args:
      shape: A tuple or list, each element of which is an int or, less often,
          `None`.
      dtype: A `dtype` object, either from NumPy or TensorFlow.

    Returns:
      A `ShapeDtype` instance whose `shape` is a tuple and `dtype` is a NumPy
      `dtype` object.
    """
    # Canonicalize shape and dtype.
    if isinstance(shape, list):
      shape = tuple(shape)
    if not isinstance(shape, tuple):
      raise TypeError('shape must be tuple or list; got: {}'.format(shape))
    if isinstance(dtype, tf.DType):
      dtype = dtype.as_numpy_dtype

    self.shape = shape
    self.dtype = dtype 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def is_structure_of_integers(type_spec: computation_types.Type) -> bool:
  """Determines if `type_spec` is a structure of integers.

  Args:
    type_spec: A `computation_types.Type`.

  Returns:
    `True` iff `type_spec` is a structure of integers, otherwise `False`.
  """
  py_typecheck.check_type(type_spec, computation_types.Type)
  if type_spec.is_tensor():
    py_typecheck.check_type(type_spec.dtype, tf.DType)
    return type_spec.dtype.is_integer
  elif type_spec.is_tuple():
    return all(
        is_structure_of_integers(v)
        for _, v in anonymous_tuple.iter_elements(type_spec))
  elif type_spec.is_federated():
    return is_structure_of_integers(type_spec.member)
  else:
    return False 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def create_tensorflow_to_broadcast_scalar(
    scalar_type: tf.dtypes.DType,
    new_shape: tf.TensorShape) -> building_blocks.CompiledComputation:
  """Creates TF function broadcasting scalar to shape `new_shape`.

  Args:
    scalar_type: Instance of `tf.DType`, the type of the scalar we are looking
      to broadcast.
    new_shape: Instance of `tf.TensorShape`, the shape we wish to broadcast to.
      Must be fully defined.

  Returns:
    Instance of `building_blocks.CompiledComputation` representing
    a function declaring a scalar parameter of dtype `scalar_type`, and
    returning a tensor of this same dtype and shape `new_shape`, with the same
    value in each entry as its scalar argument.

  Raises:
    TypeError: If the types of the arguments do not match the declared arg
    types.
    ValueError: If `new_shape` is not fully defined.
  """
  proto = tensorflow_computation_factory.create_broadcast_scalar_to_shape(
      scalar_type, new_shape)
  return building_blocks.CompiledComputation(proto) 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def open_detached(
    ciphertext: Ciphertext,
    mac: Mac,
    nonce: Nonce,
    public_sender: PublicKey,
    secretkey_receiver: SecretKey,
    plaintext_dtype: tf.DType,
) -> tf.Tensor:
    plaintext = sodium_module.sodium_easy_box_open_detached(
        ciphertext.raw,
        mac.raw,
        nonce.raw,
        public_sender.raw,
        secretkey_receiver.raw,
        plaintext_dtype,
    )
    return plaintext 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def register_inputs(self) -> None:
        feedables = set.union(*[ex.feedables for ex in self.model.runners])
        if self.train_mode:
            feedables |= set.union(
                *[ex.feedables for ex in self.model.trainers])

        # collect input shapes and types
        input_types = {}  # type: Dict[str, tf.DType]
        input_shapes = {}  # type: Dict[str, tf.TensorShape]

        for feedable in feedables:
            input_types.update(feedable.input_types)
            input_shapes.update(feedable.input_shapes)

        dataset = {}  # type: Dict[str, tf.Tensor]
        for s_id, dtype in input_types.items():
            shape = input_shapes[s_id]
            dataset[s_id] = tf.placeholder(dtype, shape, s_id)

        for feedable in feedables:
            feedable.register_input(dataset)

        self.model.dataset_runner.register_input(dataset) 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def get_variable(name: str,
                 shape: ShapeSpec = None,
                 dtype: tf.DType = None,
                 initializer: Callable = None,
                 **kwargs) -> tf.Variable:
    """Get an existing variable with these parameters or create a new one.

    This is a wrapper around `tf.get_variable`. The `initializer` parameter is
    treated as a default which can be overriden by a call to
    `update_initializers`.

    This should only be called during model building.
    """
    return tf.get_variable(
        name=name, shape=shape, dtype=dtype,
        initializer=get_initializer(name, initializer),
        **kwargs) 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def __init__(self,
                 U: List[Tensor],
                 priorU: List[Distribution],
                 likelihood: Likelihood,
                 dtype: tf.DType,
                 stopCriterion,
                 phase: Phase,
                 noiseUniformity: NoiseUniformity,
                 transform: bool = False) -> None:

        # setup the model
        self.dtype = dtype
        self.__transform = transform
        self.__noiseUniformity = noiseUniformity
        self.likelihood = likelihood
        self.stopCriterion = stopCriterion
        self.postU = []  # type: List[PostU]
        for f, priorUf in enumerate(priorU):
            postUf = PostU(likelihood, priorUf, f)
            self.postU.append(postUf)

        # create or reuse the variables for the filter banks
        for f, Uf in enumerate(copy(U)):
            if transform and (f == 0):
                paramName = "{}tr".format(f)
            else:
                paramName = "{}".format(f)
            with tf.variable_scope("U", reuse=tf.AUTO_REUSE):
                UfVar = tf.get_variable(paramName,
                                        dtype=dtype,
                                        initializer=Uf)
            U[f] = UfVar
        self.__U = tuple(U)
        if phase == Phase.EM or phase == Phase.INIT:
            self.__setEm()
        elif phase == Phase.BCD:
            self.__setBcd()
        else:
            raise ValueError 

# 需要導入模塊: import tensorflow [as 別名]
# 或者: from tensorflow import DType [as 別名]
def random(cls,
               priorU: List[Distribution],
               likelihood: Likelihood,
               M: Tuple[int, ...],
               K: int,
               dtype: tf.DType,
               phase: Phase,
               stopCriterion,
               noiseUniformity: NoiseUniformity = HOMOGENEOUS,
               transform: bool = False) -> "TensorFactorisation":

        # initialize U
        dtype = tf.as_dtype(dtype)
        zero = tf.constant(0., dtype=dtype)
        one = tf.constant(1., dtype=dtype)
        normal = tf.distributions.Normal(loc=zero, scale=one)
        F = len(M)
        U = []
        for f in range(F):
            if priorU[f].nonNegative:
                UfInit = tf.abs(normal.sample(sample_shape=(K, M[f])))
            else:
                UfInit = normal.sample(sample_shape=(K, M[f]))
            U.append(UfInit)

        # instantiate
        tefa = TensorFactorisation(U=U,
                                   priorU=priorU,
                                   likelihood=likelihood,
                                   dtype=dtype,
                                   phase=phase,
                                   transform=transform,
                                   noiseUniformity=noiseUniformity,
                                   stopCriterion=stopCriterion)
        return(tefa)