Python allennlp.modules.TimeDistributed() Examples

def __init__(self, vocab            ,
                 text_field_embedder                   ,
                 encoder                ,
                 initializer                        = InitializerApplicator(),
                 regularizer                                  = None)        :
        super(SimpleTagger, self).__init__(vocab, regularizer)

        self.text_field_embedder = text_field_embedder
        self.num_classes = self.vocab.get_vocab_size(u"labels")
        self.encoder = encoder
        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(),
                               u"text field embedding dim", u"encoder input dim")
        self.metrics = {
                u"accuracy": CategoricalAccuracy(),
                u"accuracy3": CategoricalAccuracy(top_k=3)
        }

        initializer(self)

    #overrides 

def test_time_distributed_reshapes_multiple_inputs_with_pass_through_non_tensor_correctly(self):
        class FakeModule(Module):
            @overrides
            def forward(self, input_tensor, number=0, another_tensor=None):

                return input_tensor + number + another_tensor

        module = FakeModule()
        distributed_module = TimeDistributed(module)

        input_tensor1 = torch.LongTensor([[[1, 2], [3, 4]]])
        input_number = 5
        input_tensor2 = torch.LongTensor([[[4, 2], [9, 1]]])

        output = distributed_module(
            input_tensor1,
            number=input_number,
            another_tensor=input_tensor2,
            pass_through=["number"],
        )
        assert_almost_equal(output.data.numpy(), [[[10, 9], [17, 10]]]) 

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,
                 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,
                 text_field_embedder: TextFieldEmbedder,
                 seq2seq_encoder: Seq2SeqEncoder,
                 initializer: InitializerApplicator) -> None:
        super(ProLocalModel, self).__init__(vocab)

        self.text_field_embedder = text_field_embedder
        self.seq2seq_encoder = seq2seq_encoder

        self.attention_layer = \
            Attention(similarity_function=BilinearSimilarity(2 * seq2seq_encoder.get_output_dim(),
                                                             seq2seq_encoder.get_output_dim()), normalize=True)

        self.num_types = self.vocab.get_vocab_size("state_change_type_labels")
        self.aggregate_feedforward = Linear(seq2seq_encoder.get_output_dim(),
                                            self.num_types)

        self.span_metric = SpanBasedF1Measure(vocab,
                                              tag_namespace="state_change_tags")  # by default "O" is ignored in metric computation
        self.num_tags = self.vocab.get_vocab_size("state_change_tags")

        self.tag_projection_layer = TimeDistributed(Linear(self.seq2seq_encoder.get_output_dim() + 2
                                                           , self.num_tags))
        self._type_accuracy = CategoricalAccuracy()

        self.type_f1_metrics = {}
        self.type_labels_vocab = self.vocab.get_index_to_token_vocabulary("state_change_type_labels")
        for type_label in self.type_labels_vocab.values():
            self.type_f1_metrics["type_" + type_label] = F1Measure(self.vocab.get_token_index(type_label, "state_change_type_labels"))

        self._loss = torch.nn.CrossEntropyLoss()

        initializer(self) 

def test_time_distributed_works_with_multiple_inputs(self):
        module = lambda x, y: x + y
        distributed = TimeDistributed(module)
        x_input = torch.LongTensor([[[1, 2], [3, 4]]])
        y_input = torch.LongTensor([[[4, 2], [9, 1]]])
        output = distributed(x_input, y_input)
        assert_almost_equal(output.data.numpy(), [[[5, 4], [12, 5]]]) 

def test_time_distributed_reshapes_correctly(self):
        char_embedding = Embedding(2, 2)
        char_embedding.weight = Parameter(torch.FloatTensor([[.4, .4], [.5, .5]]))
        distributed_embedding = TimeDistributed(char_embedding)
        char_input = torch.LongTensor([[[1, 0], [1, 1]]])
        output = distributed_embedding(char_input)
        assert_almost_equal(output.data.numpy(),
                            [[[[.5, .5], [.4, .4]], [[.5, .5,], [.5, .5]]]]) 

def __init__(self, vocab            ,
                 text_field_embedder                   ,
                 encoder                ,
                 binary_feature_dim     ,
                 embedding_dropout        = 0.0,
                 initializer                        = InitializerApplicator(),
                 regularizer                                  = None,
                 label_smoothing        = None)        :
        super(SemanticRoleLabeler, self).__init__(vocab, regularizer)

        self.text_field_embedder = text_field_embedder
        self.num_classes = self.vocab.get_vocab_size(u"labels")

        # For the span based evaluation, we don't want to consider labels
        # for verb, because the verb index is provided to the model.
        self.span_metric = SpanBasedF1Measure(vocab, tag_namespace=u"labels", ignore_classes=[u"V"])

        self.encoder = encoder
        # There are exactly 2 binary features for the verb predicate embedding.
        self.binary_feature_embedding = Embedding(2, binary_feature_dim)
        self.tag_projection_layer = TimeDistributed(Linear(self.encoder.get_output_dim(),
                                                           self.num_classes))
        self.embedding_dropout = Dropout(p=embedding_dropout)
        self._label_smoothing = label_smoothing

        check_dimensions_match(text_field_embedder.get_output_dim() + binary_feature_dim,
                               encoder.get_input_dim(),
                               u"text embedding dim + verb indicator embedding dim",
                               u"encoder input dim")
        initializer(self) 

def __init__(self, vocab            ,
                 text_field_embedder                   ,
                 attend_feedforward             ,
                 similarity_function                    ,
                 compare_feedforward             ,
                 aggregate_feedforward             ,
                 premise_encoder                           = None,
                 hypothesis_encoder                           = None,
                 initializer                        = InitializerApplicator(),
                 regularizer                                  = None)        :
        super(DecomposableAttention, self).__init__(vocab, regularizer)

        self._text_field_embedder = text_field_embedder
        self._attend_feedforward = TimeDistributed(attend_feedforward)
        self._matrix_attention = LegacyMatrixAttention(similarity_function)
        self._compare_feedforward = TimeDistributed(compare_feedforward)
        self._aggregate_feedforward = aggregate_feedforward
        self._premise_encoder = premise_encoder
        self._hypothesis_encoder = hypothesis_encoder or premise_encoder

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

        check_dimensions_match(text_field_embedder.get_output_dim(), attend_feedforward.get_input_dim(),
                               u"text field embedding dim", u"attend feedforward input dim")
        check_dimensions_match(aggregate_feedforward.get_output_dim(), self._num_labels,
                               u"final output dimension", u"number of labels")

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

        initializer(self) 

def test_time_distributed_works_with_multiple_inputs(self):
        module = lambda x, y: x + y
        distributed = TimeDistributed(module)
        x_input = torch.LongTensor([[[1, 2], [3, 4]]])
        y_input = torch.LongTensor([[[4, 2], [9, 1]]])
        output = distributed(x_input, y_input)
        assert_almost_equal(output.data.numpy(), [[[5, 4], [12, 5]]]) 

def test_time_distributed_reshapes_positional_kwarg_correctly(self):
        char_embedding = Embedding(2, 2)
        char_embedding.weight = Parameter(torch.FloatTensor([[0.4, 0.4], [0.5, 0.5]]))
        distributed_embedding = TimeDistributed(char_embedding)
        char_input = torch.LongTensor([[[1, 0], [1, 1]]])
        output = distributed_embedding(input=char_input)
        assert_almost_equal(
            output.data.numpy(), [[[[0.5, 0.5], [0.4, 0.4]], [[0.5, 0.5], [0.5, 0.5]]]]
        ) 

def test_time_distributed_reshapes_named_arg_correctly(self):
        char_embedding = Embedding(2, 2)
        char_embedding.weight = Parameter(torch.FloatTensor([[0.4, 0.4], [0.5, 0.5]]))
        distributed_embedding = TimeDistributed(char_embedding)
        char_input = torch.LongTensor([[[1, 0], [1, 1]]])
        output = distributed_embedding(char_input)
        assert_almost_equal(
            output.data.numpy(), [[[[0.5, 0.5], [0.4, 0.4]], [[0.5, 0.5], [0.5, 0.5]]]]
        ) 

def forward(
        self, tokens: torch.Tensor, mask: torch.BoolTensor = None, num_wrapping_dims: int = 0
    ):
        pooler = self.pooler
        for _ in range(num_wrapping_dims):
            from allennlp.modules import TimeDistributed

            pooler = TimeDistributed(pooler)
        pooled = pooler(tokens)
        pooled = self._dropout(pooled)
        return pooled 

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,
                 text_field_embedder: TextFieldEmbedder,
                 encoder: Seq2SeqEncoder,
                 label_namespace: str = "labels",
                 constraint_type: str = None,
                 include_start_end_transitions: bool = True,
                 dropout: float = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None) -> None:
        super().__init__(vocab, regularizer)

        self.label_namespace = label_namespace
        self.text_field_embedder = text_field_embedder
        self.num_tags = self.vocab.get_vocab_size(label_namespace)
        self.encoder = encoder
        if dropout:
            self.dropout = torch.nn.Dropout(dropout)
        else:
            self.dropout = None
        self.tag_projection_layer = TimeDistributed(Linear(self.encoder.get_output_dim(),
                                                           self.num_tags))

        if constraint_type is not None:
            labels = self.vocab.get_index_to_token_vocabulary(label_namespace)
            constraints = allowed_transitions(constraint_type, labels)
        else:
            constraints = None

        self.crf = ConditionalRandomField(
                self.num_tags, constraints,
                include_start_end_transitions=include_start_end_transitions
        )

        self.span_metric = SpanBasedF1Measure(vocab,
                                              tag_namespace=label_namespace,
                                              label_encoding=constraint_type or "BIO")

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

def __init__(self,
                 vocab: Vocabulary,
                 task: str,
                 encoder: Seq2SeqEncoder,
                 label_smoothing: float = 0.0,
                 dropout: float = 0.0,
                 adaptive: bool = False,
                 features: List[str] = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None) -> None:
        super(TagDecoder, self).__init__(vocab, regularizer)

        self.task = task
        self.encoder = encoder
        self.output_dim = encoder.get_output_dim()
        self.label_smoothing = label_smoothing
        self.num_classes = self.vocab.get_vocab_size(task)
        self.adaptive = adaptive
        self.features = features if features else []

        self.metrics = {
            "acc": CategoricalAccuracy(),
            # "acc3": CategoricalAccuracy(top_k=3)
        }

        if self.adaptive:
            # TODO
            adaptive_cutoffs = [round(self.num_classes / 15), 3 * round(self.num_classes / 15)]
            self.task_output = AdaptiveLogSoftmaxWithLoss(self.output_dim,
                                                          self.num_classes,
                                                          cutoffs=adaptive_cutoffs,
                                                          div_value=4.0)
        else:
            self.task_output = TimeDistributed(Linear(self.output_dim, self.num_classes))

        self.feature_outputs = torch.nn.ModuleDict()
        self.features_metrics = {}
        for feature in self.features:
            self.feature_outputs[feature] = TimeDistributed(Linear(self.output_dim,
                                                                   vocab.get_vocab_size(feature)))
            self.features_metrics[feature] = {
                "acc": CategoricalAccuracy(),
            }

        initializer(self) 

def __init__(self,
                 vocab            ,
                 text_field_embedder                   ,
                 span_extractor               ,
                 encoder                ,
                 feedforward              = None,
                 pos_tag_embedding            = None,
                 initializer                        = InitializerApplicator(),
                 regularizer                                  = None,
                 evalb_directory_path      = DEFAULT_EVALB_DIR)        :
        super(SpanConstituencyParser, self).__init__(vocab, regularizer)

        self.text_field_embedder = text_field_embedder
        self.span_extractor = span_extractor
        self.num_classes = self.vocab.get_vocab_size(u"labels")
        self.encoder = encoder
        self.feedforward_layer = TimeDistributed(feedforward) if feedforward else None
        self.pos_tag_embedding = pos_tag_embedding or None
        if feedforward is not None:
            output_dim = feedforward.get_output_dim()
        else:
            output_dim = span_extractor.get_output_dim()

        self.tag_projection_layer = TimeDistributed(Linear(output_dim, self.num_classes))

        representation_dim = text_field_embedder.get_output_dim()
        if pos_tag_embedding is not None:
            representation_dim += pos_tag_embedding.get_output_dim()
        check_dimensions_match(representation_dim,
                               encoder.get_input_dim(),
                               u"representation dim (tokens + optional POS tags)",
                               u"encoder input dim")
        check_dimensions_match(encoder.get_output_dim(),
                               span_extractor.get_input_dim(),
                               u"encoder input dim",
                               u"span extractor input dim")
        if feedforward is not None:
            check_dimensions_match(span_extractor.get_output_dim(),
                                   feedforward.get_input_dim(),
                                   u"span extractor output dim",
                                   u"feedforward input dim")

        self.tag_accuracy = CategoricalAccuracy()

        if evalb_directory_path is not None:
            self._evalb_score = EvalbBracketingScorer(evalb_directory_path)
        else:
            self._evalb_score = None
        initializer(self)

    #overrides 

def __init__(self, vocab            ,
                 text_field_embedder                   ,
                 encoder                ,
                 label_namespace      = u"labels",
                 constraint_type      = None,
                 feedforward              = None,
                 include_start_end_transitions       = True,
                 dropout        = None,
                 verbose_metrics       = False,
                 initializer                        = InitializerApplicator(),
                 regularizer                                  = None)        :
        super(CrfTagger, self).__init__(vocab, regularizer)

        self.label_namespace = label_namespace
        self.text_field_embedder = text_field_embedder
        self.num_tags = self.vocab.get_vocab_size(label_namespace)
        self.encoder = encoder
        self._verbose_metrics = verbose_metrics
        if dropout:
            self.dropout = torch.nn.Dropout(dropout)
        else:
            self.dropout = None
        self._feedforward = feedforward

        if feedforward is not None:
            output_dim = feedforward.get_output_dim()
        else:
            output_dim = self.encoder.get_output_dim()
        self.tag_projection_layer = TimeDistributed(Linear(output_dim,
                                                           self.num_tags))

        if constraint_type is not None:
            labels = self.vocab.get_index_to_token_vocabulary(label_namespace)
            constraints = allowed_transitions(constraint_type, labels)
        else:
            constraints = None

        self.crf = ConditionalRandomField(
                self.num_tags, constraints,
                include_start_end_transitions=include_start_end_transitions
        )

        self.span_metric = SpanBasedF1Measure(vocab,
                                              tag_namespace=label_namespace,
                                              label_encoding=constraint_type or u"BIO")


        check_dimensions_match(text_field_embedder.get_output_dim(), encoder.get_input_dim(),
                               u"text field embedding dim", u"encoder input dim")
        if feedforward is not None:
            check_dimensions_match(encoder.get_output_dim(), feedforward.get_input_dim(),
                                   u"encoder output dim", u"feedforward input dim")
        initializer(self)

    #overrides 

def __init__(self,
                 vocab            ,
                 question_embedder                   ,
                 action_embedding_dim     ,
                 encoder                ,
                 entity_encoder                ,
                 max_decoding_steps     ,
                 use_neighbor_similarity_for_linking       = False,
                 dropout        = 0.0,
                 num_linking_features      = 10,
                 rule_namespace      = u'rule_labels',
                 tables_directory      = u'/wikitables/')        :
        super(WikiTablesSemanticParser, self).__init__(vocab)
        self._question_embedder = question_embedder
        self._encoder = encoder
        self._entity_encoder = TimeDistributed(entity_encoder)
        self._max_decoding_steps = max_decoding_steps
        self._use_neighbor_similarity_for_linking = use_neighbor_similarity_for_linking
        if dropout > 0:
            self._dropout = torch.nn.Dropout(p=dropout)
        else:
            self._dropout = lambda x: x
        self._rule_namespace = rule_namespace
        self._denotation_accuracy = WikiTablesAccuracy(tables_directory)
        self._action_sequence_accuracy = Average()
        self._has_logical_form = Average()

        self._action_padding_index = -1  # the padding value used by IndexField
        num_actions = vocab.get_vocab_size(self._rule_namespace)
        self._action_embedder = Embedding(num_embeddings=num_actions, embedding_dim=action_embedding_dim)
        self._output_action_embedder = Embedding(num_embeddings=num_actions, embedding_dim=action_embedding_dim)
        self._action_biases = Embedding(num_embeddings=num_actions, embedding_dim=1)

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

        check_dimensions_match(entity_encoder.get_output_dim(), question_embedder.get_output_dim(),
                               u"entity word average embedding dim", u"question embedding dim")

        self._num_entity_types = 4  # TODO(mattg): get this in a more principled way somehow?
        self._num_start_types = 5  # TODO(mattg): get this in a more principled way somehow?
        self._embedding_dim = question_embedder.get_output_dim()
        self._type_params = torch.nn.Linear(self._num_entity_types, self._embedding_dim)
        self._neighbor_params = torch.nn.Linear(self._embedding_dim, self._embedding_dim)

        if num_linking_features > 0:
            self._linking_params = torch.nn.Linear(num_linking_features, 1)
        else:
            self._linking_params = None

        if self._use_neighbor_similarity_for_linking:
            self._question_entity_params = torch.nn.Linear(1, 1)
            self._question_neighbor_params = torch.nn.Linear(1, 1)
        else:
            self._question_entity_params = None
            self._question_neighbor_params = None 

def __init__(self,
                 vocab            ,
                 text_field_embedder                   ,
                 context_layer                ,
                 mention_feedforward             ,
                 antecedent_feedforward             ,
                 feature_size     ,
                 max_span_width     ,
                 spans_per_word       ,
                 max_antecedents     ,
                 lexical_dropout        = 0.2,
                 initializer                        = InitializerApplicator(),
                 regularizer                                  = None)        :
        super(CoreferenceResolver, self).__init__(vocab, regularizer)

        self._text_field_embedder = text_field_embedder
        self._context_layer = context_layer
        self._antecedent_feedforward = TimeDistributed(antecedent_feedforward)
        feedforward_scorer = torch.nn.Sequential(
                TimeDistributed(mention_feedforward),
                TimeDistributed(torch.nn.Linear(mention_feedforward.get_output_dim(), 1)))
        self._mention_pruner = SpanPruner(feedforward_scorer)
        self._antecedent_scorer = TimeDistributed(torch.nn.Linear(antecedent_feedforward.get_output_dim(), 1))

        self._endpoint_span_extractor = EndpointSpanExtractor(context_layer.get_output_dim(),
                                                              combination=u"x,y",
                                                              num_width_embeddings=max_span_width,
                                                              span_width_embedding_dim=feature_size,
                                                              bucket_widths=False)
        self._attentive_span_extractor = SelfAttentiveSpanExtractor(input_dim=text_field_embedder.get_output_dim())

        # 10 possible distance buckets.
        self._num_distance_buckets = 10
        self._distance_embedding = Embedding(self._num_distance_buckets, feature_size)

        self._max_span_width = max_span_width
        self._spans_per_word = spans_per_word
        self._max_antecedents = max_antecedents

        self._mention_recall = MentionRecall()
        self._conll_coref_scores = ConllCorefScores()
        if lexical_dropout > 0:
            self._lexical_dropout = torch.nn.Dropout(p=lexical_dropout)
        else:
            self._lexical_dropout = lambda x: x
        initializer(self)

    #overrides 

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,
                 span_encoder: Seq2SeqEncoder,
                 reasoning_encoder: Seq2SeqEncoder,
                 input_dropout: float = 0.3,
                 hidden_dim_maxpool: int = 1024,
                 class_embs: bool=True,
                 reasoning_use_obj: bool=True,
                 reasoning_use_answer: bool=True,
                 reasoning_use_question: bool=True,
                 pool_reasoning: bool = True,
                 pool_answer: bool = True,
                 pool_question: bool = False,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 ):
        super(AttentionQA, self).__init__(vocab)

        self.detector = SimpleDetector(pretrained=True, average_pool=True, semantic=class_embs, final_dim=512)
        ###################################################################################################

        self.rnn_input_dropout = TimeDistributed(InputVariationalDropout(input_dropout)) if input_dropout > 0 else None

        self.span_encoder = TimeDistributed(span_encoder)
        self.reasoning_encoder = TimeDistributed(reasoning_encoder)

        self.span_attention = BilinearMatrixAttention(
            matrix_1_dim=span_encoder.get_output_dim(),
            matrix_2_dim=span_encoder.get_output_dim(),
        )

        self.obj_attention = BilinearMatrixAttention(
            matrix_1_dim=span_encoder.get_output_dim(),
            matrix_2_dim=self.detector.final_dim,
        )

        self.reasoning_use_obj = reasoning_use_obj
        self.reasoning_use_answer = reasoning_use_answer
        self.reasoning_use_question = reasoning_use_question
        self.pool_reasoning = pool_reasoning
        self.pool_answer = pool_answer
        self.pool_question = pool_question
        dim = sum([d for d, to_pool in [(reasoning_encoder.get_output_dim(), self.pool_reasoning),
                                        (span_encoder.get_output_dim(), self.pool_answer),
                                        (span_encoder.get_output_dim(), self.pool_question)] if to_pool])

        self.final_mlp = torch.nn.Sequential(
            torch.nn.Dropout(input_dropout, inplace=False),
            torch.nn.Linear(dim, hidden_dim_maxpool),
            torch.nn.ReLU(inplace=True),
            torch.nn.Dropout(input_dropout, inplace=False),
            torch.nn.Linear(hidden_dim_maxpool, 1),
        )
        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,
                 text_field_embedder: TextFieldEmbedder,
                 attend_feedforward: FeedForward,
                 similarity_function: SimilarityFunction,
                 compare_feedforward: FeedForward,
                 aggregate_feedforward: FeedForward,
                 premise_encoder: Optional[Seq2SeqEncoder] = None,
                 hypothesis_encoder: Optional[Seq2SeqEncoder] = None,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 regularizer: Optional[RegularizerApplicator] = None,
                 preload_path: Optional[str] = None) -> None:
        super(DecomposableAttention, self).__init__(vocab, regularizer)

        self._text_field_embedder = text_field_embedder
        self._attend_feedforward = TimeDistributed(attend_feedforward)
        self._matrix_attention = MatrixAttention(similarity_function)
        self._compare_feedforward = TimeDistributed(compare_feedforward)
        self._aggregate_feedforward = aggregate_feedforward
        self._premise_encoder = premise_encoder
        self._hypothesis_encoder = hypothesis_encoder or premise_encoder

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

        check_dimensions_match(text_field_embedder.get_output_dim(), attend_feedforward.get_input_dim(),
                               "text field embedding dim", "attend feedforward input dim")
        # check_dimensions_match(aggregate_feedforward.get_output_dim(), self._num_labels,
        #                        "final output dimension", "number of labels")

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

        initializer(self)

        # Do we want to initialize with the SNLI stuff? let's say yes.
        # 'snli-decomposable-attention/weights.th'
        if preload_path is not None:
            logger.info("Preloading!")
            preload = torch.load(preload_path)
            own_state = self.state_dict()
            for name, param in preload.items():
                if name not in own_state:
                    logger.info("Unexpected key {} in state_dict with size {}".format(name, param.size()))
                elif param.size() == own_state[name].size():
                    own_state[name].copy_(param)
                else:
                    logger.info("Network has {} with size {}, ckpt has {}".format(name,
                                                                            own_state[name].size(),
                                                                            param.size()))

                missing = set(own_state.keys()) - set(preload.keys())
                if len(missing) > 0:
                    logger.info("We couldn't find {}".format(','.join(missing))) 

def __init__(self,
                 vocab: Vocabulary,
                 span_encoder: Seq2SeqEncoder,
                 reasoning_encoder: Seq2SeqEncoder,
                 input_dropout: float = 0.3,
                 hidden_dim_maxpool: int = 1024,
                 class_embs: bool = True,
                 reasoning_use_obj: bool = True,
                 reasoning_use_answer: bool = True,
                 reasoning_use_question: bool = True,
                 pool_reasoning: bool = True,
                 pool_answer: bool = True,
                 pool_question: bool = False,
                 initializer: InitializerApplicator = InitializerApplicator(),
                 ):
        super(HGL_Model, self).__init__(vocab)

        self.detector = SimpleDetector(pretrained=True, average_pool=True, semantic=class_embs, final_dim=512)
        ###################################################################################################
        self.rnn_input_dropout = TimeDistributed(InputVariationalDropout(input_dropout)) if input_dropout > 0 else None
        self.span_encoder = TimeDistributed(span_encoder)
        self.reasoning_encoder = TimeDistributed(reasoning_encoder)

        self.Graph_reasoning = Graph_reasoning(512)

        self.QAHG = BilinearMatrixAttention(
            matrix_1_dim=span_encoder.get_output_dim(),
            matrix_2_dim=span_encoder.get_output_dim(),
        )

        self.VAHG = BilinearMatrixAttention(
            matrix_1_dim=span_encoder.get_output_dim(),
            matrix_2_dim=self.detector.final_dim,
        )

        self.reasoning_use_obj = reasoning_use_obj
        self.reasoning_use_answer = reasoning_use_answer
        self.reasoning_use_question = reasoning_use_question
        self.pool_reasoning = pool_reasoning
        self.pool_answer = pool_answer
        self.pool_question = pool_question
        dim = sum([d for d, to_pool in [(reasoning_encoder.get_output_dim(), self.pool_reasoning),
                                        (span_encoder.get_output_dim(), self.pool_answer),
                                        (span_encoder.get_output_dim(), self.pool_question)] if to_pool])

        self.final_mlp = torch.nn.Sequential(
            torch.nn.Dropout(input_dropout, inplace=False),
            torch.nn.Linear(dim, hidden_dim_maxpool),
            torch.nn.ReLU(inplace=True),
            torch.nn.Dropout(input_dropout, inplace=False),
            torch.nn.Linear(hidden_dim_maxpool, 1),
        )
        self._accuracy = CategoricalAccuracy()
        self._loss = torch.nn.CrossEntropyLoss()
        initializer(self)