Python rnn._transpose_batch_time方法代码示例

# 需要导入模块: from tensorflow.python.ops import rnn [as 别名]
# 或者: from tensorflow.python.ops.rnn import _transpose_batch_time [as 别名]
def transpose_batch_time(inputs):
    """Transposes inputs between time-major and batch-major.

    Args:
        inputs: A Tensor of shape `[batch_size, max_time, ...]` (batch-major)
            or `[max_time, batch_size, ...]` (time-major), or a (possibly
            nested) tuple of such elements.

    Returns:
        A (possibly nested tuple of) Tensor with transposed batch and
        time dimensions of inputs.
    """
    flat_input = nest.flatten(inputs)
    flat_input = [ops.convert_to_tensor(input_) for input_ in flat_input]
    # pylint: disable=protected-access
    flat_input = [rnn._transpose_batch_time(input_) for input_ in flat_input]
    return nest.pack_sequence_as(structure=inputs, flat_sequence=flat_input) 

# 需要导入模块: from tensorflow.python.ops import rnn [as 别名]
# 或者: from tensorflow.python.ops.rnn import _transpose_batch_time [as 别名]
def _mask_sequences_tensor(sequence,
                           sequence_length,
                           dtype=None,
                           time_major=False,
                           tensor_rank=2):
    """Masks out sequence entries that are beyond the respective sequence
    lengths. Masks along the time dimension.

    Args:
        sequence: A Tensor of sequence values.

            If `time_major=False` (default), this must be a Tensor of shape:
                `[batch_size, max_time, d_2, ..., d_rank]`, where the rank of
                the Tensor is specified with :attr:`tensor_rank`.

            If `time_major=True`, this must be a Tensor of shape:
                `[max_time, batch_size, d_2, ..., d_rank].`
        sequence_length: A Tensor of shape `[batch_size]`. Time steps beyond
            the respective sequence lengths will be made zero.
        dtype: Type of :attr:`sequence`. If `None`, inferred from
            :attr:`sequence` automatically.
        time_major (bool): The shape format of the inputs. If `True`,
            :attr:`sequence` must have shape
            `[max_time, batch_size, d_2, ..., d_rank]`.
            If `False` (default), :attr:`sequence` must have
            shape `[batch_size, max_time, d_2, ..., d_rank]`.
        tensor_rank (int): The number of dimensions of :attr:`sequence`.
            Default is 2, i.e., :attr:`sequence` is a 2D Tensor consisting
            of batch and time dimensions.

    Returns:
        The masked sequence, i.e., a Tensor of the same shape as
        :attr:`sequence` but with masked-out entries (set to zero).
    """
    if tensor_rank is None:
        tensor_rank = 2
    if tensor_rank < 2:
        raise ValueError(
            "tensor_rank must be > 2. Got tensor_rank = {}".format(tensor_rank))
    if time_major:
        sequence = rnn._transpose_batch_time(sequence)
    max_time = tf.to_int32(tf.shape(sequence)[1])
    if dtype is None:
        dtype = sequence.dtype
    mask = tf.sequence_mask(
        tf.to_int32(sequence_length), max_time, dtype=dtype)
    for _ in range(2, tensor_rank):
        mask = tf.expand_dims(mask, axis=-1)
    sequence = sequence * mask
    if time_major:
        sequence = rnn._transpose_batch_time(sequence)
    return sequence 

# 需要导入模块: from tensorflow.python.ops import rnn [as 别名]
# 或者: from tensorflow.python.ops.rnn import _transpose_batch_time [as 别名]
def _mask_sequences_tensor(sequence,
                           sequence_length,
                           dtype=None,
                           time_major=False,
                           tensor_rank=2):
    """Masks out sequence entries that are beyond the respective sequence
    lengths. Masks along the time dimension.

    Args:
        sequence: A Tensor of sequence values.

            If `time_major=False` (default), this must be a Tensor of shape:
                `[batch_size, max_time, d_2, ..., d_rank]`, where the rank of
                the Tensor is specified with :attr:`tensor_rank`.

            If `time_major=True`, this must be a Tensor of shape:
                `[max_time, batch_size, d_2, ..., d_rank].`
        sequence_length: A Tensor of shape `[batch_size]`. Time steps beyond
            the respective sequence lengths will be made zero.
        dtype (dtype): Type of :attr:`sequence`. If `None`, infer from
            :attr:`sequence` automatically.
        time_major (bool): The shape format of the inputs. If `True`,
            :attr:`sequence` must have shape
            `[max_time, batch_size, d_2, ..., d_rank]`.
            If `False` (default), :attr:`sequence` must have
            shape `[batch_size, max_time, d_2, ..., d_rank]`.
        tensor_rank (int): The number of dimensions of :attr:`sequence`.
            Default is 2, i.e., :attr:`sequence` is a 2D Tensor consisting
            of batch and time dimensions.

    Returns:
        The masked sequence, i.e., a Tensor of the same shape as
        :attr:`sequence` but with masked-out entries (set to zero).
    """
    if tensor_rank is None:
        tensor_rank = 2
    if tensor_rank < 2:
        raise ValueError(
            "tensor_rank must be > 2. Got tensor_rank = {}".format(tensor_rank))
    if time_major:
        sequence = rnn._transpose_batch_time(sequence)
    max_time = tf.cast(tf.shape(sequence)[1], tf.int32)
    if dtype is None:
        dtype = sequence.dtype
    mask = tf.sequence_mask(
        tf.cast(sequence_length, tf.int32), max_time, dtype=dtype)
    for _ in range(2, tensor_rank):
        mask = tf.expand_dims(mask, axis=-1)
    sequence = sequence * mask
    if time_major:
        sequence = rnn._transpose_batch_time(sequence)
    return sequence