def __init__(self,
output_series: str,
decoder: Decoder) -> None:
check_argument_types()
BaseRunner.__init__(self, output_series, decoder)
self._decoder_xent = cast(Decoder, self._decoder).train_xents
def __init__(self,
output_series: str,
decoder: SequenceRegressor,
postprocess: Callable[[float], float] = None) -> None:
check_argument_types()
BaseRunner.__init__(self, output_series, decoder)
self._postprocess = postprocess
def __init__(self,
output_series: str,
decoder: Any,
normalize: bool = True,
pick_index: int = None,
pick_value: str = None) -> None:
"""Initializes the logits runner.
Args:
output_series: Name of the series produces by the runner.
decoder: A decoder having logits.
normalize: Flag whether the logits should be normalized with
softmax.
pick_index: If not None, it specifies the index of the logit or the
probability that should be on output.
pick_value: If not None, it specifies a value from the decoder's
vocabulary whose logit or probability should be on output.
"""
check_argument_types()
BaseRunner.__init__(self, output_series, decoder)
if not hasattr(self._decoder, "vocabulary"):
raise TypeError("Decoder for logits runner should have the "
"'vocabulary' attribute")
if not hasattr(self._decoder, "decoded_logits"):
raise TypeError("Decoder for logits runner should have the "
"'decoded_logits' attribute")
# TODO why these two?
if not hasattr(self._decoder, "train_loss"):
raise TypeError("Decoder for logits runner should have the "
"'train_loss' attribute")
if not hasattr(self._decoder, "runtime_loss"):
raise TypeError("Decoder for logits runner should have the "
"'runtime_loss' attribute")
self._decoder_vocab = getattr(self._decoder, "vocabulary")
self._decoder_train_loss = getattr(self._decoder, "train_loss")
self._decoder_runtime_loss = getattr(self._decoder, "runtime_loss")
self._decoder_logits = getattr(self._decoder, "decoded_logits")
if pick_index is not None and pick_value is not None:
raise ValueError("Either a pick index or a vocabulary value can "
"be specified, not both at the same time.")
self._normalize = normalize
if pick_value is not None:
if pick_value in self._decoder_vocab:
vocab_map = self._decoder_vocab.word_to_index
self._pick_index = vocab_map[pick_value]
else:
raise ValueError(
"Value '{}' is not in vocabulary of decoder '{}'".format(
pick_value, decoder.name))
else:
self._pick_index = pick_index
def __init__(self,
output_series: str,
decoder: BeamSearchDecoder,
rank: int = 1,
postprocess: Callable[[List[str]], List[str]] = None) -> None:
check_argument_types()
BaseRunner.__init__(self, output_series, decoder)
if rank < 1 or rank > decoder.beam_size:
raise ValueError(
("Rank of output hypothesis must be between 1 and the beam "
"size ({}), was {}.").format(decoder.beam_size, rank))
self._rank = rank
self._postprocess = postprocess
def __init__(self,
output_series: str,
encoder: ModelPart,
attribute: str = "output",
used_session: int = 0) -> None:
"""Initialize the representation runner.
Args:
output_series: Name of the output series with vectors.
encoder: The encoder to use.
attribute: The name of the encoder attribute that contains the
data.
used_session: Id of the TensorFlow session used in case of model
ensembles.
"""
check_argument_types()
BaseRunner.__init__(self, output_series, cast(ModelPart, encoder))
self._used_session = used_session # type: int
self._encoded = getattr(encoder, attribute) # type: tf.Tensor
def __init__(self,
output_series: str,
encoder: Stateful,
used_session: int = 0) -> None:
"""Initialize the representation runner.
Args:
output_series: Name of the output seriesi with vectors.
encoder: Used encoder.
used_session: Id of the TensorFlow session used in case of model
ensembles.
"""
check_argument_types()
if not isinstance(encoder, ModelPart):
raise TypeError("The encoder of the representation runner has to "
"be an instance of 'ModelPart'")
BaseRunner.__init__(self, output_series, cast(ModelPart, encoder))
self._used_session = used_session # type: int
self._encoded = encoder.output # type: Stateful
def __init__(self, output_series: str, decoder: BeamSearchDecoder,
attention: BaseAttention) -> None:
check_argument_types()
BaseRunner.__init__(self, output_series, decoder)