Python allennlp.modules.TextFieldEmbedder() Examples

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 verbose_metrics: bool = False,
                 dropout: float = 0.2,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None,
                 ) -> None:
        super(TextClassifier, self).__init__(vocab, regularizer)

        self.text_field_embedder = text_field_embedder
        self.dropout = torch.nn.Dropout(dropout)
        self.num_classes = self.vocab.get_vocab_size("labels")
        self.classifier_feedforward = torch.nn.Linear(self.text_field_embedder.get_output_dim()  , self.num_classes)

        self.label_accuracy = CategoricalAccuracy()
        self.label_f1_metrics = {}

        self.verbose_metrics = verbose_metrics

        for i in range(self.num_classes):
            self.label_f1_metrics[vocab.get_token_from_index(index=i, namespace="labels")] = F1Measure(positive_label=i)
        self.loss = torch.nn.CrossEntropyLoss()

        initializer(self) 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 text_encoder: Seq2SeqEncoder,
                 classifier_feedforward: FeedForward,
                 verbose_metrics: False,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None,
                 ) -> None:
        super(TextClassifier, self).__init__(vocab, regularizer)

        self.text_field_embedder = text_field_embedder
        self.num_classes = self.vocab.get_vocab_size("labels")
        self.text_encoder = text_encoder
        self.classifier_feedforward = classifier_feedforward
        self.prediction_layer = torch.nn.Linear(self.classifier_feedforward.get_output_dim()  , self.num_classes)

        self.label_accuracy = CategoricalAccuracy()
        self.label_f1_metrics = {}

        self.verbose_metrics = verbose_metrics

        for i in range(self.num_classes):
            self.label_f1_metrics[vocab.get_token_from_index(index=i, namespace="labels")] = F1Measure(positive_label=i)
        self.loss = torch.nn.CrossEntropyLoss()

        self.pool = lambda text, mask: util.get_final_encoder_states(text, mask, bidirectional=True)

        initializer(self) 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 calculate_span_f1: bool = None,
                 label_encoding: Optional[str] = None,
                 label_namespace: str = "labels",
                 verbose_metrics: bool = False,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None) -> None:
        super(SimpleTagger, self).__init__(vocab, regularizer)

        self.label_namespace = label_namespace
        self.text_field_embedder = text_field_embedder
        self.num_classes = self.vocab.get_vocab_size(label_namespace)
        self.encoder = encoder
        self._verbose_metrics = verbose_metrics
        self.tag_projection_layer = TimeDistributed(Linear(self.encoder.get_output_dim(),
                                                           self.num_classes))

        check_dimensions_match(text_field_embedder.get_output_dim(), encoder.get_input_dim(),
                               "text field embedding dim", "encoder input dim")

        # We keep calculate_span_f1 as a constructor argument for API consistency with
        # the CrfTagger, even it is redundant in this class
        # (label_encoding serves the same purpose).
        if calculate_span_f1 and not label_encoding:
            raise ConfigurationError("calculate_span_f1 is True, but "
                                     "no label_encoding was specified.")
        self.metrics = {
            "accuracy": CategoricalAccuracy(),
            "accuracy3": CategoricalAccuracy(top_k=3)
        }

        if calculate_span_f1 or label_encoding:
            self._f1_metric = SpanBasedF1Measure(vocab,
                                                 tag_namespace=label_namespace,
                                                 label_encoding=label_encoding)
        else:
            self._f1_metric = None

        initializer(self) 

def __init__(self,
                 vocab: Vocabulary,
                 input_embedder: TextFieldEmbedder,
                 encoder: Encoder = None,
                 dropout: float = None,
                 initializer: InitializerApplicator = InitializerApplicator()
                ) -> None:
        """
        Parameters
        ----------
        vocab: `Vocabulary`
            vocab to use
        input_embedder: `TextFieldEmbedder`
            generic embedder of tokens
        encoder: `Encoder`, optional (default = None)
            Seq2Vec or Seq2Seq Encoder wrapper. If no encoder is provided,
            assume that the input is a bag of word counts, for linear classification.
        dropout: `float`, optional (default = None)
            if set, will apply dropout to output of encoder.
        initializer: `InitializerApplicator`
            generic initializer
        """
        super().__init__(vocab)
        self._input_embedder = input_embedder
        if dropout:
            self._dropout = torch.nn.Dropout(dropout)
        else:
            self._dropout = None
        self._encoder = encoder
        self._num_labels = vocab.get_vocab_size(namespace="labels")
        if self._encoder:
            self._clf_input_dim = self._encoder.get_output_dim()
        else:
            self._clf_input_dim = self._input_embedder.get_output_dim()
        self._classification_layer = torch.nn.Linear(self._clf_input_dim,
                                                     self._num_labels)
        self._accuracy = CategoricalAccuracy()
        self._loss = torch.nn.CrossEntropyLoss()
        initializer(self) 

def __init__(self,
                 vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 question_encoder: Seq2SeqEncoder,
                 passage_encoder: Seq2SeqEncoder,
                 pair_encoder: AttentionEncoder,
                 self_encoder: AttentionEncoder,
                 output_layer: QAOutputLayer,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None,
                 share_encoder: bool = False):

        super().__init__(vocab, regularizer)
        self.text_field_embedder = text_field_embedder
        self.question_encoder = question_encoder
        self.passage_encoder = passage_encoder
        self.pair_encoder = pair_encoder
        self.self_encoder = self_encoder
        self.output_layer = output_layer

        self._span_start_accuracy = CategoricalAccuracy()
        self._span_end_accuracy = CategoricalAccuracy()
        self._span_accuracy = BooleanAccuracy()
        self._squad_metrics = SquadEmAndF1()
        self.share_encoder = share_encoder
        self.loss = torch.nn.CrossEntropyLoss()
        initializer(self) 

def __init__(self,
                 vocab: Vocabulary,
                 source_text_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 decoder: SeqDecoder,
                 tied_source_embedder_key: Optional[str] = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None) -> None:

        super(CustomComposedSeq2Seq, self).__init__(vocab, regularizer)

        self._source_text_embedder = source_text_embedder
        self._encoder = encoder
        self._decoder = decoder

        if self._encoder.get_output_dim() != self._decoder.get_output_dim():
            raise ConfigurationError(f"Encoder output dimension {self._encoder.get_output_dim()} should be"
                                     f" equal to decoder dimension {self._decoder.get_output_dim()}.")
        if tied_source_embedder_key:
            if not isinstance(self._source_text_embedder, BasicTextFieldEmbedder):
                raise ConfigurationError("Unable to tie embeddings,"
                                         "Source text embedder is not an instance of `BasicTextFieldEmbedder`.")
            source_embedder = self._source_text_embedder._token_embedders[tied_source_embedder_key]
            if not isinstance(source_embedder, Embedding):
                raise ConfigurationError("Unable to tie embeddings,"
                                         "Selected source embedder is not an instance of `Embedding`.")
            if source_embedder.get_output_dim() != self._decoder.target_embedder.get_output_dim():
                raise ConfigurationError(f"Output Dimensions mismatch between"
                                         f"source embedder and target embedder.")
            self._source_text_embedder._token_embedders[tied_source_embedder_key] = self._decoder.target_embedder
        initializer(self) 

def __init__(self,
                 vocab: Vocabulary,
                 source_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 attention: Attention,
                 beam_size: int,
                 max_decoding_steps: int,
                 target_embedding_dim: int = None,
                 copy_token: str = "@COPY@",
                 source_namespace: str = "tokens",
                 target_namespace: str = "target_tokens",
                 tensor_based_metric: Metric = None,
                 token_based_metric: Metric = None,
                 tie_embeddings: bool = False) -> None:
        target_embedding_dim = target_embedding_dim or source_embedder.get_output_dim()
        CopyNetSeq2Seq.__init__(
            self,
            vocab,
            source_embedder,
            encoder,
            attention,
            beam_size,
            max_decoding_steps,
            target_embedding_dim,
            copy_token,
            source_namespace,
            target_namespace,
            tensor_based_metric,
            token_based_metric
        )
        self._tie_embeddings = tie_embeddings

        if self._tie_embeddings:
            assert source_namespace == target_namespace
            assert "token_embedder_tokens" in dict(self._source_embedder.named_children())
            source_token_embedder = dict(self._source_embedder.named_children())["token_embedder_tokens"]
            self._target_embedder.weight = source_token_embedder.weight

        if tensor_based_metric is None:
            self._tensor_based_metric = None 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 phrase_layer: Seq2SeqEncoder,
                 projected_layer: Seq2SeqEncoder,
                 flow_layer: Seq2SeqEncoder,
                 contextual_passage: Seq2SeqEncoder,
                 contextual_question: Seq2SeqEncoder,
                 dropout: float = 0.2,
                 regularizer: Optional[RegularizerApplicator] = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 ):

        super(MultiGranularityHierarchicalAttentionFusionNetworks, self).__init__(vocab, regularizer)
        self._text_field_embedder = text_field_embedder
        self._phrase_layer = phrase_layer
        self._encoding_dim = self._phrase_layer.get_output_dim()
        self.projected_layer = torch.nn.Linear(self._encoding_dim + 1024, self._encoding_dim)
        self.fuse = FusionLayer(self._encoding_dim)
        self.projected_lstm = projected_layer
        self.flow = flow_layer
        self.contextual_layer_p = contextual_passage
        self.contextual_layer_q = contextual_question
        self.linear_self_align = torch.nn.Linear(self._encoding_dim, 1)
        self.bilinear_layer_s = BilinearSeqAtt(self._encoding_dim, self._encoding_dim)
        self.bilinear_layer_e = BilinearSeqAtt(self._encoding_dim, self._encoding_dim)
        self.yesno_predictor = torch.nn.Linear(self._encoding_dim, 3)
        self.relu = torch.nn.ReLU()

        self._max_span_length = 30

        self._span_start_accuracy = CategoricalAccuracy()
        self._span_end_accuracy = CategoricalAccuracy()
        self._span_accuracy = BooleanAccuracy()
        self._squad_metrics = SquadEmAndF1()
        self._span_yesno_accuracy = CategoricalAccuracy()
        self._official_f1 = Average()
        self._variational_dropout = InputVariationalDropout(dropout)

        self._loss = torch.nn.CrossEntropyLoss()
        initializer(self) 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 projection_feedforward: FeedForward,
                 inference_encoder: Seq2SeqEncoder,
                 output_feedforward: FeedForward,
                 output_logit: FeedForward,
                 final_feedforward: FeedForward,
                 coverage_loss: CoverageLoss,
                 similarity_function: SimilarityFunction = DotProductSimilarity(),
                 dropout: float = 0.5,
                 contextualize_pair_comparators: bool = False,
                 pair_context_encoder: Seq2SeqEncoder = None,
                 pair_feedforward: FeedForward = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None) -> None:
        super().__init__(vocab=vocab,
                         text_field_embedder=text_field_embedder,
                         encoder=encoder,
                         similarity_function=similarity_function,
                         projection_feedforward=projection_feedforward,
                         inference_encoder=inference_encoder,
                         output_feedforward=output_feedforward,
                         output_logit=output_logit,
                         final_feedforward=final_feedforward,
                         coverage_loss=coverage_loss,
                         contextualize_pair_comparators=contextualize_pair_comparators,
                         pair_context_encoder=pair_context_encoder,
                         pair_feedforward=pair_feedforward,
                         dropout=dropout,
                         initializer=initializer,
                         regularizer=regularizer)
        self._ignore_index = -1
        self._answer_loss = torch.nn.CrossEntropyLoss(ignore_index=self._ignore_index)
        self._coverage_loss = coverage_loss

        self._accuracy = CategoricalAccuracy()
        self._entailment_f1 = F1Measure(self._label2idx["entailment"]) 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 projection_feedforward: FeedForward,
                 inference_encoder: Seq2SeqEncoder,
                 output_feedforward: FeedForward,
                 output_logit: FeedForward,
                 final_feedforward: FeedForward,
                 coverage_loss: CoverageLoss,
                 similarity_function: SimilarityFunction = DotProductSimilarity(),
                 dropout: float = 0.5,
                 contextualize_pair_comparators: bool = False,
                 pair_context_encoder: Seq2SeqEncoder = None,
                 pair_feedforward: FeedForward = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None) -> None:
    	# Need to send it verbatim because otherwise FromParams doesn't work appropriately.
        super().__init__(vocab=vocab,
                         text_field_embedder=text_field_embedder,
                         encoder=encoder,
                         similarity_function=similarity_function,
                         projection_feedforward=projection_feedforward,
                         inference_encoder=inference_encoder,
                         output_feedforward=output_feedforward,
                         output_logit=output_logit,
                         final_feedforward=final_feedforward,
                         contextualize_pair_comparators=contextualize_pair_comparators,
                         coverage_loss=coverage_loss,
                         pair_context_encoder=pair_context_encoder,
                         pair_feedforward=pair_feedforward,
                         dropout=dropout,
                         initializer=initializer,
                         regularizer=regularizer)
        self._answer_loss = torch.nn.CrossEntropyLoss()

        self._accuracy = CategoricalAccuracy() 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 include_start_end_transitions: bool = True,
                 dropout: Optional[float] = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None) -> None:
        super().__init__(vocab, regularizer)

        self.label_namespace = 'labels'
        self.num_tags = self.vocab.get_vocab_size(self.label_namespace)

        # encode text
        self.text_field_embedder = text_field_embedder
        self.encoder = encoder
        self.dropout = torch.nn.Dropout(dropout) if dropout else None

        # crf
        output_dim = self.encoder.get_output_dim()
        self.tag_projection_layer = TimeDistributed(Linear(output_dim, self.num_tags))
        self.crf = ConditionalRandomField(self.num_tags, constraints=None, include_start_end_transitions=include_start_end_transitions)

        self.metrics = {
            "accuracy": CategoricalAccuracy(),
            "accuracy3": CategoricalAccuracy(top_k=3)
        }
        for index, label in self.vocab.get_index_to_token_vocabulary(self.label_namespace).items():
            self.metrics['F1_' + label] = F1Measure(positive_label=index)

        initializer(self) 

def __init__(
        self,
        vocab: Vocabulary,
        sentence_embedder: TextFieldEmbedder,
        action_embedding_dim: int,
        encoder: Seq2SeqEncoder,
        attention: Attention,
        decoder_beam_search: BeamSearch,
        max_decoding_steps: int,
        dropout: float = 0.0,
    ) -> None:
        super(NlvrDirectSemanticParser, self).__init__(
            vocab=vocab,
            sentence_embedder=sentence_embedder,
            action_embedding_dim=action_embedding_dim,
            encoder=encoder,
            dropout=dropout,
        )
        self._decoder_trainer = MaximumMarginalLikelihood()
        self._decoder_step = BasicTransitionFunction(
            encoder_output_dim=self._encoder.get_output_dim(),
            action_embedding_dim=action_embedding_dim,
            input_attention=attention,
            activation=Activation.by_name("tanh")(),
            add_action_bias=False,
            dropout=dropout,
        )
        self._decoder_beam_search = decoder_beam_search
        self._max_decoding_steps = max_decoding_steps
        self._action_padding_index = -1 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 predictor_dropout=0.0,
                 labels_namespace: str = "labels",
                 detect_namespace: str = "d_tags",
                 verbose_metrics: bool = False,
                 label_smoothing: float = 0.0,
                 confidence: float = 0.0,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None) -> None:
        super(Seq2Labels, self).__init__(vocab, regularizer)

        self.label_namespaces = [labels_namespace,
                                 detect_namespace]
        self.text_field_embedder = text_field_embedder
        self.num_labels_classes = self.vocab.get_vocab_size(labels_namespace)
        self.num_detect_classes = self.vocab.get_vocab_size(detect_namespace)
        self.label_smoothing = label_smoothing
        self.confidence = confidence
        self.incorr_index = self.vocab.get_token_index("INCORRECT",
                                                       namespace=detect_namespace)

        self._verbose_metrics = verbose_metrics
        self.predictor_dropout = TimeDistributed(torch.nn.Dropout(predictor_dropout))

        self.tag_labels_projection_layer = TimeDistributed(
            Linear(text_field_embedder._token_embedders['bert'].get_output_dim(), self.num_labels_classes))

        self.tag_detect_projection_layer = TimeDistributed(
            Linear(text_field_embedder._token_embedders['bert'].get_output_dim(), self.num_detect_classes))

        self.metrics = {"accuracy": CategoricalAccuracy()}

        initializer(self) 

def __init__(
        self,
        vocab: Vocabulary,
        sentence_embedder: TextFieldEmbedder,
        action_embedding_dim: int,
        encoder: Seq2SeqEncoder,
        dropout: float = 0.0,
        rule_namespace: str = "rule_labels",
    ) -> None:
        super(NlvrSemanticParser, self).__init__(vocab=vocab)

        self._sentence_embedder = sentence_embedder
        self._denotation_accuracy = Average()
        self._consistency = Average()
        self._encoder = encoder
        if dropout > 0:
            self._dropout = torch.nn.Dropout(p=dropout)
        else:
            self._dropout = lambda x: x
        self._rule_namespace = rule_namespace

        self._action_embedder = Embedding(
            num_embeddings=vocab.get_vocab_size(self._rule_namespace),
            embedding_dim=action_embedding_dim,
        )

        # This is what we pass as input in the first step of decoding, when we don't have a
        # previous action.
        self._first_action_embedding = torch.nn.Parameter(torch.FloatTensor(action_embedding_dim))
        torch.nn.init.normal_(self._first_action_embedding) 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 phrase_layer: Seq2SeqEncoder,
                 projected_layer: Seq2SeqEncoder,
                 contextual_passage: Seq2SeqEncoder,
                 contextual_question: Seq2SeqEncoder,
                 dropout: float = 0.2,
                 regularizer: Optional[RegularizerApplicator] = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 ):

        super(MultiGranularityHierarchicalAttentionFusionNetworks, self).__init__(vocab, regularizer)
        self._text_field_embedder = text_field_embedder
        self._phrase_layer = phrase_layer
        self._encoding_dim = self._phrase_layer.get_output_dim()
        self.projected_layer = torch.nn.Linear(self._encoding_dim + 1024, self._encoding_dim)
        self.fuse_p = FusionLayer(self._encoding_dim)
        self.fuse_q = FusionLayer(self._encoding_dim)
        self.fuse_s = FusionLayer(self._encoding_dim)
        self.projected_lstm = projected_layer
        self.contextual_layer_p = contextual_passage
        self.contextual_layer_q = contextual_question
        self.linear_self_align = torch.nn.Linear(self._encoding_dim, 1)
        # self.bilinear_self_align = BilinearSelfAlign(self._encoding_dim)
        # self._self_attention = LinearMatrixAttention(self._encoding_dim, self._encoding_dim, 'x,y,x*y')
        self._self_attention = BilinearMatrixAttention(self._encoding_dim, self._encoding_dim)
        self.bilinear_layer_s = BilinearSeqAtt(self._encoding_dim, self._encoding_dim)
        self.bilinear_layer_e = BilinearSeqAtt(self._encoding_dim, self._encoding_dim)
        self.yesno_predictor = torch.nn.Linear(self._encoding_dim, 3)
        self.relu = torch.nn.ReLU()

        self._max_span_length = 30

        self._span_start_accuracy = CategoricalAccuracy()
        self._span_end_accuracy = CategoricalAccuracy()
        self._span_accuracy = BooleanAccuracy()
        self._squad_metrics = SquadEmAndF1()
        self._span_yesno_accuracy = CategoricalAccuracy()
        self._official_f1 = Average()
        self._variational_dropout = InputVariationalDropout(dropout)

        self._loss = torch.nn.CrossEntropyLoss()
        initializer(self) 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 sentence_encoder: Seq2VecEncoder,
                 classifier_feedforward: FeedForward,
                 label_weight: Dict[str, float] = None,
                 use_label_distribution: bool = False,
                 image_classification_ratio: float = 0.0,
                 decay_every_i_step=100000,
                 decay_ratio=0.8,
                 instance_count=100000,
                 max_epoch=10,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None
                 ) -> None:
        super(BasicClassifier, self).__init__(vocab, regularizer)

        self.text_field_embedder = text_field_embedder
        self.num_classes = self.vocab.get_vocab_size("labels")
        self.sentence_encoder = sentence_encoder
        self.classifier_feedforward = classifier_feedforward

        if text_field_embedder.get_output_dim() != sentence_encoder.get_input_dim():
            raise ConfigurationError("The output dimension of the text_field_embedder must match the "
                                     "input dimension of the title_encoder. Found {} and {}, "
                                     "respectively.".format(text_field_embedder.get_output_dim(),
                                                            sentence_encoder.get_input_dim()))
        self.metrics = {
                "accuracy": CategoricalAccuracy(),
                "cnn_loss": Average()
        }
        if not use_label_distribution:
            self.loss = torch.nn.CrossEntropyLoss()
        else:
            self.loss = torch.nn.CrossEntropyLoss()
        self.image_classification_ratio = image_classification_ratio
        self.decay_every_i_step = decay_every_i_step
        self.decay_ratio = decay_ratio
        self.training_step = 0
        self.current_ratio = image_classification_ratio
        self.total_steps = max_epoch*instance_count//64
        self.step_every_epoch = instance_count // 64

        print("每个epoch的step数量", self.step_every_epoch)

        initializer(self) 

def forward(self,  # type: ignore
                tokens: Dict[str, torch.LongTensor],
                tags: torch.LongTensor = None,
                metadata: List[Dict[str, Any]] = None) -> Dict[str, torch.Tensor]:
        # pylint: disable=arguments-differ
        """
        Parameters
        ----------
        tokens : Dict[str, torch.LongTensor], required
            The output of ``TextField.as_array()``, which should typically be passed directly to a
            ``TextFieldEmbedder``. This output is a dictionary mapping keys to ``TokenIndexer``
            tensors.  At its most basic, using a ``SingleIdTokenIndexer`` this is: ``{"tokens":
            Tensor(batch_size, num_tokens)}``. This dictionary will have the same keys as were used
            for the ``TokenIndexers`` when you created the ``TextField`` representing your
            sequence.  The dictionary is designed to be passed directly to a ``TextFieldEmbedder``,
            which knows how to combine different word representations into a single vector per
            token in your input.
        tags : torch.LongTensor, optional (default = None)
            A torch tensor representing the sequence of integer gold class labels of shape
            ``(batch_size, num_tokens)``.
        metadata : ``List[Dict[str, Any]]``, optional, (default = None)
            metadata containing the original words in the sentence to be tagged under a 'words' key.

        Returns
        -------
        An output dictionary consisting of:
        logits : torch.FloatTensor
            A tensor of shape ``(batch_size, num_tokens, tag_vocab_size)`` representing
            unnormalised log probabilities of the tag classes.
        class_probabilities : torch.FloatTensor
            A tensor of shape ``(batch_size, num_tokens, tag_vocab_size)`` representing
            a distribution of the tag classes per word.
        loss : torch.FloatTensor, optional
            A scalar loss to be optimised.

        """
        embedded_text_input = self.text_field_embedder(tokens)
        batch_size, sequence_length, _ = embedded_text_input.size()
        mask = get_text_field_mask(tokens)
        encoded_text = self.encoder(embedded_text_input, mask)

        logits = self.tag_projection_layer(encoded_text)
        reshaped_log_probs = logits.view(-1, self.num_classes)
        class_probabilities = F.softmax(reshaped_log_probs, dim=-1).view([batch_size,
                                                                          sequence_length,
                                                                          self.num_classes])

        output_dict = {"logits": logits, "class_probabilities": class_probabilities}

        if tags is not None:
            loss = sequence_cross_entropy_with_logits(logits, tags, mask)
            for metric in self.metrics.values():
                metric(logits, tags, mask.float())
            if self._f1_metric is not None:
                self._f1_metric(logits, tags, mask.float())
            output_dict["loss"] = loss

        if metadata is not None:
            output_dict["words"] = [x["words"] for x in metadata]
        return output_dict 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 aggregate_feedforward: FeedForward,
                 premise_encoder: Optional[Seq2SeqEncoder] = None,
                 hypothesis_encoder: Optional[Seq2SeqEncoder] = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 aggregate_premise: Optional[str] = "max",
                 aggregate_hypothesis: Optional[str] = "max",
                 embeddings_dropout_value: Optional[float] = 0.0,
                 share_encoders: Optional[bool] = False) -> None:
        super(StackedNNAggregateCustom, self).__init__(vocab)

        self._text_field_embedder = text_field_embedder
        if embeddings_dropout_value > 0.0:
            self._embeddings_dropout = torch.nn.Dropout(p=embeddings_dropout_value)
        else:
            self._embeddings_dropout = lambda x: x

        self._aggregate_feedforward = aggregate_feedforward
        self._premise_encoder = premise_encoder
        self._hypothesis_encoder = hypothesis_encoder

        self._premise_aggregate = aggregate_premise
        self._hypothesis_aggregate = aggregate_hypothesis

        self._num_labels = vocab.get_vocab_size(namespace="labels")

        premise_output_dim = self._text_field_embedder.get_output_dim()
        if self._premise_encoder is not None:
            premise_output_dim = self._premise_encoder.get_output_dim()

        hypothesis_output_dim = self._text_field_embedder.get_output_dim()
        if self._hypothesis_encoder is not None:
            hypothesis_output_dim = self._hypothesis_encoder.get_output_dim()

        if premise_output_dim != hypothesis_output_dim:
            raise ConfigurationError("Output dimension of the premise_encoder (dim: {}), "
                                     "plus hypothesis_encoder (dim: {})"
                                     "must match! "
                                     .format(premise_output_dim,
                                             hypothesis_output_dim))

        if premise_output_dim * 4 != \
                aggregate_feedforward.get_input_dim():
            raise ConfigurationError("The output of aggregate_feedforward input dim ({2})  "
                                     "should be {3} = 4 x {0} ({1} = premise_output_dim == hypothesis_output_dim)!"
                                     .format(premise_output_dim,
                                             hypothesis_output_dim,
                                             aggregate_feedforward.get_input_dim(),
                                             4 * premise_output_dim))

        if aggregate_feedforward.get_output_dim() != self._num_labels:
            raise ConfigurationError("Final output dimension (%d) must equal num labels (%d)" %
                                     (aggregate_feedforward.get_output_dim(), self._num_labels))

        self._accuracy = CategoricalAccuracy()
        self._loss = torch.nn.CrossEntropyLoss()

        initializer(self) 

def __init__(
        self,
        vocab: Vocabulary,
        text_field_embedder: TextFieldEmbedder,
        encoder: Seq2SeqEncoder,
        calculate_span_f1: bool = None,
        label_encoding: Optional[str] = None,
        label_namespace: str = "labels",
        verbose_metrics: bool = False,
        initializer: InitializerApplicator = InitializerApplicator(),
        **kwargs,
    ) -> None:
        super().__init__(vocab, **kwargs)

        self.label_namespace = label_namespace
        self.text_field_embedder = text_field_embedder
        self.num_classes = self.vocab.get_vocab_size(label_namespace)
        self.encoder = encoder
        self._verbose_metrics = verbose_metrics
        self.tag_projection_layer = TimeDistributed(
            Linear(self.encoder.get_output_dim(), self.num_classes)
        )

        check_dimensions_match(
            text_field_embedder.get_output_dim(),
            encoder.get_input_dim(),
            "text field embedding dim",
            "encoder input dim",
        )

        self.metrics = {
            "accuracy": CategoricalAccuracy(),
            "accuracy3": CategoricalAccuracy(top_k=3),
        }

        # We keep calculate_span_f1 as a constructor argument for API consistency with
        # the CrfTagger, even it is redundant in this class
        # (label_encoding serves the same purpose).
        if calculate_span_f1 is None:
            calculate_span_f1 = label_encoding is not None

        self.calculate_span_f1 = calculate_span_f1
        if calculate_span_f1:
            if not label_encoding:
                raise ConfigurationError(
                    "calculate_span_f1 is True, but no label_encoding was specified."
                )
            self._f1_metric = SpanBasedF1Measure(
                vocab, tag_namespace=label_namespace, label_encoding=label_encoding
            )
        else:
            self._f1_metric = None

        initializer(self) 

def __init__(
        self,
        vocab: Vocabulary,
        text_field_embedder: TextFieldEmbedder,
        seq2vec_encoder: Seq2VecEncoder,
        seq2seq_encoder: Seq2SeqEncoder = None,
        feedforward: Optional[FeedForward] = None,
        dropout: float = None,
        num_labels: int = None,
        label_namespace: str = "labels",
        namespace: str = "tokens",
        initializer: InitializerApplicator = InitializerApplicator(),
        **kwargs,
    ) -> None:

        super().__init__(vocab, **kwargs)
        self._text_field_embedder = text_field_embedder

        if seq2seq_encoder:
            self._seq2seq_encoder = seq2seq_encoder
        else:
            self._seq2seq_encoder = None

        self._seq2vec_encoder = seq2vec_encoder
        self._feedforward = feedforward
        if feedforward is not None:
            self._classifier_input_dim = self._feedforward.get_output_dim()
        else:
            self._classifier_input_dim = self._seq2vec_encoder.get_output_dim()

        if dropout:
            self._dropout = torch.nn.Dropout(dropout)
        else:
            self._dropout = None
        self._label_namespace = label_namespace
        self._namespace = namespace

        if num_labels:
            self._num_labels = num_labels
        else:
            self._num_labels = vocab.get_vocab_size(namespace=self._label_namespace)
        self._classification_layer = torch.nn.Linear(self._classifier_input_dim, self._num_labels)
        self._accuracy = CategoricalAccuracy()
        self._loss = torch.nn.CrossEntropyLoss()
        initializer(self) 

def __init__(self,
                 vocab: Vocabulary,
                 source_embedder: TextFieldEmbedder,
                 sentence_encoder: Seq2VecEncoder,
                 sentence_accumulator: Seq2SeqEncoder,
                 use_salience: bool,
                 use_pos_embedding: bool,
                 use_output_bias: bool,
                 use_novelty: bool,
                 dropout: float = 0.3,
                 pos_embedding_num: int = 50,
                 pos_embedding_size: int = 128) -> None:
        super(SummaRuNNer, self).__init__(vocab)

        self._source_embedder = source_embedder

        self._sentence_encoder = sentence_encoder
        self._se_output_dim = self._sentence_encoder.get_output_dim()

        self._sentence_accumulator = sentence_accumulator
        self._h_sentence_dim = self._sentence_accumulator.get_output_dim()

        self._dropout_layer = Dropout(dropout)

        self._content_projection_layer = Linear(self._h_sentence_dim, 1)

        self._use_salience = use_salience
        if use_salience:
            self._document_linear_layer = Linear(self._h_sentence_dim, self._h_sentence_dim, bias=True)
            self._salience_linear_layer = Linear(self._h_sentence_dim, self._h_sentence_dim, bias=False)

        self._use_pos_embedding = use_pos_embedding
        if use_pos_embedding:
            self._pos_embedding_num = pos_embedding_num
            self._pos_embedding_size = pos_embedding_size
            self._pos_embedding_layer = Embedding(pos_embedding_num, pos_embedding_size)
            self._pos_projection_layer = Linear(pos_embedding_size, 1)

        self._use_output_bias = use_output_bias
        if use_output_bias:
            self._output_bias = Parameter(torch.zeros(1).uniform_(-0.1,0.1), requires_grad=True)

        self._use_novelty = use_novelty
        if use_novelty:
            self._novelty_linear_layer = Linear(self._h_sentence_dim, self._h_sentence_dim, bias=False) 

def __init__(self,
                 vocab: Vocabulary,
                 source_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 max_decoding_steps: int,
                 attention: Attention = None,
                 beam_size: int = None,
                 target_namespace: str = "tokens",
                 target_embedding_dim: int = None,
                 scheduled_sampling_ratio: float = 0.,
                 use_projection: bool = False,
                 projection_dim: int = None,
                 tie_embeddings: bool = False) -> None:
        super(Seq2Seq, self).__init__(
            vocab,
            source_embedder,
            encoder,
            max_decoding_steps,
            attention,
            None,
            beam_size,
            target_namespace,
            target_embedding_dim,
            scheduled_sampling_ratio
        )
        use_projection = use_projection or projection_dim is not None

        self._tie_embeddings = tie_embeddings

        if self._tie_embeddings:
            assert "token_embedder_tokens" in dict(self._source_embedder.named_children())
            source_token_embedder = dict(self._source_embedder.named_children())["token_embedder_tokens"]
            self._target_embedder.weight = source_token_embedder.weight

        num_classes = self.vocab.get_vocab_size(self._target_namespace)
        self._use_projection = use_projection
        if self._use_projection:
            self._projection_dim = projection_dim or self._source_embedder.get_output_dim()
            self._hidden_projection_layer = Linear(self._decoder_output_dim, self._projection_dim)
            self._output_projection_layer = Linear(self._projection_dim, num_classes)
        else:
            self._output_projection_layer = Linear(self._decoder_output_dim, num_classes)
        self._bleu = False 

def __init__(self,
                 vocab: Vocabulary,
                 tasks: List[str],
                 text_field_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 decoders: Dict[str, Model],
                 post_encoder_embedder: TextFieldEmbedder = None,
                 dropout: float = 0.0,
                 word_dropout: float = 0.0,
                 mix_embedding: int = None,
                 layer_dropout: int = 0.0,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None) -> None:
        super(UdifyModel, self).__init__(vocab, regularizer)

        self.tasks = sorted(tasks)
        self.vocab = vocab
        self.bert_vocab = BertTokenizer.from_pretrained("config/archive/bert-base-multilingual-cased/vocab.txt").vocab
        self.text_field_embedder = text_field_embedder
        self.post_encoder_embedder = post_encoder_embedder
        self.shared_encoder = encoder
        self.word_dropout = word_dropout
        self.dropout = torch.nn.Dropout(p=dropout)
        self.decoders = torch.nn.ModuleDict(decoders)

        if mix_embedding:
            self.scalar_mix = torch.nn.ModuleDict({
                task: ScalarMixWithDropout(mix_embedding,
                                           do_layer_norm=False,
                                           dropout=layer_dropout)
                for task in self.decoders
            })
        else:
            self.scalar_mix = None

        self.metrics = {}

        for task in self.tasks:
            if task not in self.decoders:
                raise ConfigurationError(f"Task {task} has no corresponding decoder. Make sure their names match.")

        check_dimensions_match(text_field_embedder.get_output_dim(), encoder.get_input_dim(),
                               "text field embedding dim", "encoder input dim")

        initializer(self)
        self._count_params() 

def __init__(self, vocab: Vocabulary,
                 elmo_embedder: TextFieldEmbedder,
                 tokens_embedder: TextFieldEmbedder,
                 features_embedder: TextFieldEmbedder,
                 phrase_layer: Seq2SeqEncoder,
                 projected_layer: Seq2SeqEncoder,
                 contextual_passage: Seq2SeqEncoder,
                 contextual_question: Seq2SeqEncoder,
                 dropout: float = 0.2,
                 regularizer: Optional[RegularizerApplicator] = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 ):

        super(MultiGranularityHierarchicalAttentionFusionNetworks, self).__init__(vocab, regularizer)
        self.elmo_embedder = elmo_embedder
        self.tokens_embedder = tokens_embedder
        self.features_embedder = features_embedder
        self._phrase_layer = phrase_layer
        self._encoding_dim = self._phrase_layer.get_output_dim()
        self.projected_layer = torch.nn.Linear(self._encoding_dim + 1024, self._encoding_dim)
        self.fuse_p = FusionLayer(self._encoding_dim)
        self.fuse_q = FusionLayer(self._encoding_dim)
        self.fuse_s = FusionLayer(self._encoding_dim)
        self.projected_lstm = projected_layer
        self.contextual_layer_p = contextual_passage
        self.contextual_layer_q = contextual_question
        self.linear_self_align = torch.nn.Linear(self._encoding_dim, 1)
        # self._self_attention = LinearMatrixAttention(self._encoding_dim, self._encoding_dim, 'x,y,x*y')
        self._self_attention = BilinearMatrixAttention(self._encoding_dim, self._encoding_dim)
        self.bilinear_layer_s = BilinearSeqAtt(self._encoding_dim, self._encoding_dim)
        self.bilinear_layer_e = BilinearSeqAtt(self._encoding_dim, self._encoding_dim)
        self.yesno_predictor = FeedForward(self._encoding_dim, self._encoding_dim, 3)
        self.relu = torch.nn.ReLU()

        self._max_span_length = 30

        self._span_start_accuracy = CategoricalAccuracy()
        self._span_end_accuracy = CategoricalAccuracy()
        self._span_accuracy = BooleanAccuracy()
        self._squad_metrics = SquadEmAndF1()
        self._span_yesno_accuracy = CategoricalAccuracy()
        self._official_f1 = Average()
        self._variational_dropout = InputVariationalDropout(dropout)

        self._loss = torch.nn.CrossEntropyLoss()
        initializer(self) 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 similarity_function: SimilarityFunction,
                 projection_feedforward: FeedForward,
                 inference_encoder: Seq2SeqEncoder,
                 output_feedforward: FeedForward,
                 output_logit: FeedForward,
                 final_feedforward: FeedForward,
                 coverage_loss: CoverageLoss = None,
                 contextualize_pair_comparators: bool = False,
                 pair_context_encoder: Seq2SeqEncoder = None,
                 pair_feedforward: FeedForward = None,
                 optimize_coverage_for: List = ["entailment", "neutral"],
                 dropout: float = 0.5,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None) -> None:
        super().__init__(vocab, regularizer)
        self._label2idx = self.vocab.get_token_to_index_vocabulary('labels')

        self._text_field_embedder = text_field_embedder

        self._entailment_comparator_layer_1 = EsimComparatorLayer1(encoder, dropout)
        self._entailment_comparator_layer_2 = EsimComparatorLayer2(similarity_function)

        self._td_entailment_comparator_layer_1 = TimeDistributed(self._entailment_comparator_layer_1)
        self._td_entailment_comparator_layer_2 = TimeDistributed(self._entailment_comparator_layer_2)

        self._entailment_comparator_layer_3plus_local = EsimComparatorLayer3Plus(projection_feedforward, inference_encoder,
                                                                                 output_feedforward, dropout)
        self._td_entailment_comparator_layer_3plus_local = EachOutputTimeDistributed(self._entailment_comparator_layer_3plus_local)

        self._entailment_comparator_layer_3plus_global = copy.deepcopy(self._entailment_comparator_layer_3plus_local)

        self._contextualize_pair_comparators = contextualize_pair_comparators

        if not self._contextualize_pair_comparators:
            self._output_logit = output_logit
            self._td_output_logit = TimeDistributed(self._output_logit)

        self._final_feedforward = final_feedforward
        self._td_final_feedforward = TimeDistributed(final_feedforward)

        linear = torch.nn.Linear(2*self._entailment_comparator_layer_3plus_local.get_output_dim(),
                                 self._final_feedforward.get_input_dim())
        self._local_global_projection = torch.nn.Sequential(linear, torch.nn.ReLU())

        if self._contextualize_pair_comparators:
            self._pair_context_encoder = pair_context_encoder
            self._td_pair_feedforward = TimeDistributed(pair_feedforward)

        self._coverage_loss = coverage_loss

        # Do not apply initializer. If you do, make sure it doesn't reinitialize transferred parameters. 

def forward(self,  # type: ignore
                tokens                             ,
                tags                   = None,
                metadata                       = None)                           :
        # pylint: disable=arguments-differ
        u"""
        Parameters
        ----------
        tokens : Dict[str, torch.LongTensor], required
            The output of ``TextField.as_array()``, which should typically be passed directly to a
            ``TextFieldEmbedder``. This output is a dictionary mapping keys to ``TokenIndexer``
            tensors.  At its most basic, using a ``SingleIdTokenIndexer`` this is: ``{"tokens":
            Tensor(batch_size, num_tokens)}``. This dictionary will have the same keys as were used
            for the ``TokenIndexers`` when you created the ``TextField`` representing your
            sequence.  The dictionary is designed to be passed directly to a ``TextFieldEmbedder``,
            which knows how to combine different word representations into a single vector per
            token in your input.
        tags : torch.LongTensor, optional (default = None)
            A torch tensor representing the sequence of integer gold class labels of shape
            ``(batch_size, num_tokens)``.
        metadata : ``List[Dict[str, Any]]``, optional, (default = None)
            metadata containg the original words in the sentence to be tagged under a 'words' key.

        Returns
        -------
        An output dictionary consisting of:
        logits : torch.FloatTensor
            A tensor of shape ``(batch_size, num_tokens, tag_vocab_size)`` representing
            unnormalised log probabilities of the tag classes.
        class_probabilities : torch.FloatTensor
            A tensor of shape ``(batch_size, num_tokens, tag_vocab_size)`` representing
            a distribution of the tag classes per word.
        loss : torch.FloatTensor, optional
            A scalar loss to be optimised.

        """
        embedded_text_input = self.text_field_embedder(tokens)
        batch_size, sequence_length, _ = embedded_text_input.size()
        mask = get_text_field_mask(tokens)
        encoded_text = self.encoder(embedded_text_input, mask)

        logits = self.tag_projection_layer(encoded_text)
        reshaped_log_probs = logits.view(-1, self.num_classes)
        class_probabilities = F.softmax(reshaped_log_probs, dim=-1).view([batch_size,
                                                                          sequence_length,
                                                                          self.num_classes])

        output_dict = {u"logits": logits, u"class_probabilities": class_probabilities}

        if tags is not None:
            loss = sequence_cross_entropy_with_logits(logits, tags, mask)
            for metric in list(self.metrics.values()):
                metric(logits, tags, mask.float())
            output_dict[u"loss"] = loss

        if metadata is not None:
            output_dict[u"words"] = [x[u"words"] for x in metadata]
        return output_dict

    #overrides 

def __init__(self, vocab: Vocabulary,
                 encoder_keys: List[str],
                 mask_key: str,
                 pair2vec_config_file: str,
                 pair2vec_model_file: str,
                 text_field_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 similarity_function: SimilarityFunction,
                 projection_feedforward: FeedForward,
                 inference_encoder: Seq2SeqEncoder,
                 output_feedforward: FeedForward,
                 output_logit: FeedForward,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 dropout: float = 0.5,
                 pair2vec_dropout: float = 0.0,
                 bidirectional_pair2vec: bool = True,
                 regularizer: Optional[RegularizerApplicator] = None) -> None:
        super().__init__(vocab, regularizer)
        self._vocab = vocab
        self.pair2vec = util.get_pair2vec(pair2vec_config_file, pair2vec_model_file)
        self._encoder_keys = encoder_keys
        self._mask_key = mask_key
        self._text_field_embedder = text_field_embedder
        self._projection_feedforward = projection_feedforward
        self._encoder = encoder
        from allennlp.modules.matrix_attention import DotProductMatrixAttention

        self._matrix_attention = DotProductMatrixAttention()


        self._inference_encoder = inference_encoder
        self._pair2vec_dropout = torch.nn.Dropout(pair2vec_dropout)
        self._bidirectional_pair2vec = bidirectional_pair2vec

        if dropout:
            self.dropout = torch.nn.Dropout(dropout)
            self.rnn_input_dropout = VariationalDropout(dropout)
        else:
            self.dropout = None
            self.rnn_input_dropout = None

        self._output_feedforward = output_feedforward
        self._output_logit = output_logit

        self._num_labels = vocab.get_vocab_size(namespace="labels")


        self._accuracy = CategoricalAccuracy()
        self._loss = torch.nn.CrossEntropyLoss()

        initializer(self) 

def __init__(self, vocab: Vocabulary,
                 text_field_embedder: TextFieldEmbedder,
                 phrase_layer: Seq2SeqEncoder,
                 residual_encoder: Seq2SeqEncoder,
                 span_start_encoder: Seq2SeqEncoder,
                 span_end_encoder: Seq2SeqEncoder,
                 initializer: InitializerApplicator,
                 dropout: float = 0.2,
                 pair2vec_dropout: float = 0.15,
                 max_span_length: int = 30,
                 pair2vec_model_file: str = None,
                 pair2vec_config_file: str = None
                 ) -> None:
        super().__init__(vocab)
        self._max_span_length = max_span_length
        self._text_field_embedder = text_field_embedder
        self._phrase_layer = phrase_layer
        self._encoding_dim = phrase_layer.get_output_dim()

        self.pair2vec = pair2vec_util.get_pair2vec(pair2vec_config_file, pair2vec_model_file)
        self._pair2vec_dropout = torch.nn.Dropout(pair2vec_dropout)

        self._matrix_attention = LinearMatrixAttention(self._encoding_dim, self._encoding_dim, 'x,y,x*y')

        # atten_dim = self._encoding_dim * 4 + 600 if ablation_type == 'attn_over_rels' else self._encoding_dim * 4
        atten_dim = self._encoding_dim * 4 + 600
        self._merge_atten = TimeDistributed(torch.nn.Linear(atten_dim, self._encoding_dim))

        self._residual_encoder = residual_encoder

        self._self_attention = LinearMatrixAttention(self._encoding_dim, self._encoding_dim, 'x,y,x*y')

        self._merge_self_attention = TimeDistributed(torch.nn.Linear(self._encoding_dim * 3,
                                                                     self._encoding_dim))

        self._span_start_encoder = span_start_encoder
        self._span_end_encoder = span_end_encoder

        self._span_start_predictor = TimeDistributed(torch.nn.Linear(self._encoding_dim, 1))
        self._span_end_predictor = TimeDistributed(torch.nn.Linear(self._encoding_dim, 1))
        self._squad_metrics = SquadEmAndF1()
        initializer(self)

        self._span_start_accuracy = CategoricalAccuracy()
        self._span_end_accuracy = CategoricalAccuracy()
        self._official_em = Average()
        self._official_f1 = Average()

        self._span_accuracy = BooleanAccuracy()
        self._variational_dropout = InputVariationalDropout(dropout) 

def __init__(
        self,
        vocab: Vocabulary,
        utterance_embedder: TextFieldEmbedder,
        action_embedding_dim: int,
        encoder: Seq2SeqEncoder,
        decoder_beam_search: BeamSearch,
        max_decoding_steps: int,
        input_attention: Attention,
        add_action_bias: bool = True,
        dropout: float = 0.0,
        initializer: InitializerApplicator = InitializerApplicator(),
        regularizer: Optional[RegularizerApplicator] = None,
    ) -> None:
        super().__init__(vocab, regularizer)

        self._utterance_embedder = utterance_embedder
        self._encoder = encoder
        self._max_decoding_steps = max_decoding_steps
        self._add_action_bias = add_action_bias
        self._dropout = torch.nn.Dropout(p=dropout)

        self._exact_match = Average()
        self._valid_sql_query = Average()
        self._action_similarity = Average()
        self._denotation_accuracy = Average()

        # the padding value used by IndexField
        self._action_padding_index = -1
        num_actions = vocab.get_vocab_size("rule_labels")
        input_action_dim = action_embedding_dim
        if self._add_action_bias:
            input_action_dim += 1
        self._action_embedder = Embedding(
            num_embeddings=num_actions, embedding_dim=input_action_dim
        )
        self._output_action_embedder = Embedding(
            num_embeddings=num_actions, embedding_dim=action_embedding_dim
        )

        # This is what we pass as input in the first step of decoding, when we don't have a
        # previous action, or a previous utterance attention.
        self._first_action_embedding = torch.nn.Parameter(torch.FloatTensor(action_embedding_dim))
        self._first_attended_utterance = torch.nn.Parameter(
            torch.FloatTensor(encoder.get_output_dim())
        )
        torch.nn.init.normal_(self._first_action_embedding)
        torch.nn.init.normal_(self._first_attended_utterance)

        self._beam_search = decoder_beam_search
        self._decoder_trainer = MaximumMarginalLikelihood(beam_size=1)
        self._transition_function = BasicTransitionFunction(
            encoder_output_dim=self._encoder.get_output_dim(),
            action_embedding_dim=action_embedding_dim,
            input_attention=input_attention,
            add_action_bias=self._add_action_bias,
            dropout=dropout,
        )
        initializer(self)