INSTRUCTION
stringlengths
1
46.3k
RESPONSE
stringlengths
75
80.2k
Performs the functionality associated with dict.pop(key), along with checking for returned dictionaries, replacing them with Param objects with an updated history. If ``key`` is not present in the dictionary, and no default was specified, we raise a ``ConfigurationError``, instead of the typica...
def pop(self, key: str, default: Any = DEFAULT) -> Any: """ Performs the functionality associated with dict.pop(key), along with checking for returned dictionaries, replacing them with Param objects with an updated history. If ``key`` is not present in the dictionary, and no default was...
Performs a pop and coerces to an int.
def pop_int(self, key: str, default: Any = DEFAULT) -> int: """ Performs a pop and coerces to an int. """ value = self.pop(key, default) if value is None: return None else: return int(value)
Performs a pop and coerces to a float.
def pop_float(self, key: str, default: Any = DEFAULT) -> float: """ Performs a pop and coerces to a float. """ value = self.pop(key, default) if value is None: return None else: return float(value)
Performs a pop and coerces to a bool.
def pop_bool(self, key: str, default: Any = DEFAULT) -> bool: """ Performs a pop and coerces to a bool. """ value = self.pop(key, default) if value is None: return None elif isinstance(value, bool): return value elif value == "true": ...
Performs the functionality associated with dict.get(key) but also checks for returned dicts and returns a Params object in their place with an updated history.
def get(self, key: str, default: Any = DEFAULT): """ Performs the functionality associated with dict.get(key) but also checks for returned dicts and returns a Params object in their place with an updated history. """ if default is self.DEFAULT: try: va...
Gets the value of ``key`` in the ``params`` dictionary, ensuring that the value is one of the given choices. Note that this `pops` the key from params, modifying the dictionary, consistent with how parameters are processed in this codebase. Parameters ---------- key: str ...
def pop_choice(self, key: str, choices: List[Any], default_to_first_choice: bool = False) -> Any: """ Gets the value of ``key`` in the ``params`` dictionary, ensuring that the value is one of the given choices. Note that this `pops` the key from params, modifying the dictionary, consiste...
Sometimes we need to just represent the parameters as a dict, for instance when we pass them to PyTorch code. Parameters ---------- quiet: bool, optional (default = False) Whether to log the parameters before returning them as a dict. infer_type_and_cast : bool, opti...
def as_dict(self, quiet: bool = False, infer_type_and_cast: bool = False): """ Sometimes we need to just represent the parameters as a dict, for instance when we pass them to PyTorch code. Parameters ---------- quiet: bool, optional (default = False) Whether ...
Returns the parameters of a flat dictionary from keys to values. Nested structure is collapsed with periods.
def as_flat_dict(self): """ Returns the parameters of a flat dictionary from keys to values. Nested structure is collapsed with periods. """ flat_params = {} def recurse(parameters, path): for key, value in parameters.items(): newpath = path + ...
Raises a ``ConfigurationError`` if ``self.params`` is not empty. We take ``class_name`` as an argument so that the error message gives some idea of where an error happened, if there was one. ``class_name`` should be the name of the `calling` class, the one that got extra parameters (if there a...
def assert_empty(self, class_name: str): """ Raises a ``ConfigurationError`` if ``self.params`` is not empty. We take ``class_name`` as an argument so that the error message gives some idea of where an error happened, if there was one. ``class_name`` should be the name of the `calling`...
Load a `Params` object from a configuration file. Parameters ---------- params_file : ``str`` The path to the configuration file to load. params_overrides : ``str``, optional A dict of overrides that can be applied to final object. e.g. {"model.embedd...
def from_file(params_file: str, params_overrides: str = "", ext_vars: dict = None) -> 'Params': """ Load a `Params` object from a configuration file. Parameters ---------- params_file : ``str`` The path to the configuration file to load. params_overrides : ``...
Returns Ordered Dict of Params from list of partial order preferences. Parameters ---------- preference_orders: List[List[str]], optional ``preference_orders`` is list of partial preference orders. ["A", "B", "C"] means "A" > "B" > "C". For multiple preference_orders fir...
def as_ordered_dict(self, preference_orders: List[List[str]] = None) -> OrderedDict: """ Returns Ordered Dict of Params from list of partial order preferences. Parameters ---------- preference_orders: List[List[str]], optional ``preference_orders`` is list of partial...
Returns a hash code representing the current state of this ``Params`` object. We don't want to implement ``__hash__`` because that has deeper python implications (and this is a mutable object), but this will give you a representation of the current state.
def get_hash(self) -> str: """ Returns a hash code representing the current state of this ``Params`` object. We don't want to implement ``__hash__`` because that has deeper python implications (and this is a mutable object), but this will give you a representation of the current state. ...
Clears out the tracked metrics, but keeps the patience and should_decrease settings.
def clear(self) -> None: """ Clears out the tracked metrics, but keeps the patience and should_decrease settings. """ self._best_so_far = None self._epochs_with_no_improvement = 0 self._is_best_so_far = True self._epoch_number = 0 self.best_epoch = None
A ``Trainer`` can use this to serialize the state of the metric tracker.
def state_dict(self) -> Dict[str, Any]: """ A ``Trainer`` can use this to serialize the state of the metric tracker. """ return { "best_so_far": self._best_so_far, "patience": self._patience, "epochs_with_no_improvement": self._epochs_with_...
Record a new value of the metric and update the various things that depend on it.
def add_metric(self, metric: float) -> None: """ Record a new value of the metric and update the various things that depend on it. """ new_best = ((self._best_so_far is None) or (self._should_decrease and metric < self._best_so_far) or (not self._s...
Helper to add multiple metrics at once.
def add_metrics(self, metrics: Iterable[float]) -> None: """ Helper to add multiple metrics at once. """ for metric in metrics: self.add_metric(metric)
Returns true if improvement has stopped for long enough.
def should_stop_early(self) -> bool: """ Returns true if improvement has stopped for long enough. """ if self._patience is None: return False else: return self._epochs_with_no_improvement >= self._patience
Archive the model weights, its training configuration, and its vocabulary to `model.tar.gz`. Include the additional ``files_to_archive`` if provided. Parameters ---------- serialization_dir: ``str`` The directory where the weights and vocabulary are written out. weights: ``str``, option...
def archive_model(serialization_dir: str, weights: str = _DEFAULT_WEIGHTS, files_to_archive: Dict[str, str] = None, archive_path: str = None) -> None: """ Archive the model weights, its training configuration, and its vocabulary to `model.tar.gz`. Includ...
Instantiates an Archive from an archived `tar.gz` file. Parameters ---------- archive_file: ``str`` The archive file to load the model from. weights_file: ``str``, optional (default = None) The weights file to use. If unspecified, weights.th in the archive_file will be used. cuda_d...
def load_archive(archive_file: str, cuda_device: int = -1, overrides: str = "", weights_file: str = None) -> Archive: """ Instantiates an Archive from an archived `tar.gz` file. Parameters ---------- archive_file: ``str`` The archive file t...
This method can be used to load a module from the pretrained model archive. It is also used implicitly in FromParams based construction. So instead of using standard params to construct a module, you can instead load a pretrained module from the model archive directly. For eg, instead of using ...
def extract_module(self, path: str, freeze: bool = True) -> Module: """ This method can be used to load a module from the pretrained model archive. It is also used implicitly in FromParams based construction. So instead of using standard params to construct a module, you can instead loa...
Takes a list of possible actions and indices of decoded actions into those possible actions for a batch and returns sequences of action strings. We assume ``action_indices`` is a dict mapping batch indices to k-best decoded sequence lists.
def _get_action_strings(cls, possible_actions: List[List[ProductionRule]], action_indices: Dict[int, List[List[int]]]) -> List[List[List[str]]]: """ Takes a list of possible actions and indices of decoded actions into those possible actions ...
This method overrides ``Model.decode``, which gets called after ``Model.forward``, at test time, to finalize predictions. We only transform the action string sequences into logical forms here.
def decode(self, output_dict: Dict[str, torch.Tensor]) -> Dict[str, torch.Tensor]: """ This method overrides ``Model.decode``, which gets called after ``Model.forward``, at test time, to finalize predictions. We only transform the action string sequences into logical forms here. ...
Returns whether action history in the state evaluates to the correct denotations over all worlds. Only defined when the state is finished.
def _check_state_denotations(self, state: GrammarBasedState, worlds: List[NlvrLanguage]) -> List[bool]: """ Returns whether action history in the state evaluates to the correct denotations over all worlds. Only defined when the state is finished. """ assert state.is_finished(), "...
Start learning rate finder for given args
def find_learning_rate_from_args(args: argparse.Namespace) -> None: """ Start learning rate finder for given args """ params = Params.from_file(args.param_path, args.overrides) find_learning_rate_model(params, args.serialization_dir, start_lr=args.start_lr, ...
Runs learning rate search for given `num_batches` and saves the results in ``serialization_dir`` Parameters ---------- params : ``Params`` A parameter object specifying an AllenNLP Experiment. serialization_dir : ``str`` The directory in which to save results. start_lr: ``float`` ...
def find_learning_rate_model(params: Params, serialization_dir: str, start_lr: float = 1e-5, end_lr: float = 10, num_batches: int = 100, linear_steps: bool = False, stopping_f...
Runs training loop on the model using :class:`~allennlp.training.trainer.Trainer` increasing learning rate from ``start_lr`` to ``end_lr`` recording the losses. Parameters ---------- trainer: :class:`~allennlp.training.trainer.Trainer` start_lr: ``float`` The learning rate to start the searc...
def search_learning_rate(trainer: Trainer, start_lr: float = 1e-5, end_lr: float = 10, num_batches: int = 100, linear_steps: bool = False, stopping_factor: float = None) -> Tuple[List[float], Lis...
Exponential smoothing of values
def _smooth(values: List[float], beta: float) -> List[float]: """ Exponential smoothing of values """ avg_value = 0. smoothed = [] for i, value in enumerate(values): avg_value = beta * avg_value + (1 - beta) * value smoothed.append(avg_value / (1 - beta ** (i + 1))) return smoothed
Compute a weighted average of the ``tensors``. The input tensors an be any shape with at least two dimensions, but must all be the same shape. When ``do_layer_norm=True``, the ``mask`` is required input. If the ``tensors`` are dimensioned ``(dim_0, ..., dim_{n-1}, dim_n)``, then the ``mask``...
def forward(self, tensors: List[torch.Tensor], # pylint: disable=arguments-differ mask: torch.Tensor = None) -> torch.Tensor: """ Compute a weighted average of the ``tensors``. The input tensors an be any shape with at least two dimensions, but must all be the same shape. ...
Like :func:`predicate`, but used when some of the arguments to the function are meant to be provided by the decoder or other state, instead of from the language. For example, you might want to have a function use the decoder's attention over some input text when a terminal was predicted. That attention wo...
def predicate_with_side_args(side_arguments: List[str]) -> Callable: # pylint: disable=invalid-name """ Like :func:`predicate`, but used when some of the arguments to the function are meant to be provided by the decoder or other state, instead of from the language. For example, you might want to have ...
Given an ``nltk.Tree`` representing the syntax tree that generates a logical form, this method produces the actual (lisp-like) logical form, with all of the non-terminal symbols converted into the correct number of parentheses. This is used in the logic that converts action sequences back into logical form...
def nltk_tree_to_logical_form(tree: Tree) -> str: """ Given an ``nltk.Tree`` representing the syntax tree that generates a logical form, this method produces the actual (lisp-like) logical form, with all of the non-terminal symbols converted into the correct number of parentheses. This is used in t...
Converts a python ``Type`` (as you might get from a type annotation) into a ``PredicateType``. If the ``Type`` is callable, this will return a ``FunctionType``; otherwise, it will return a ``BasicType``. ``BasicTypes`` have a single ``name`` parameter - we typically get this from ``typ...
def get_type(type_: Type) -> 'PredicateType': """ Converts a python ``Type`` (as you might get from a type annotation) into a ``PredicateType``. If the ``Type`` is callable, this will return a ``FunctionType``; otherwise, it will return a ``BasicType``. ``BasicTypes`` have a si...
Executes a logical form, using whatever predicates you have defined.
def execute(self, logical_form: str): """Executes a logical form, using whatever predicates you have defined.""" if not hasattr(self, '_functions'): raise RuntimeError("You must call super().__init__() in your Language constructor") logical_form = logical_form.replace(",", " ") ...
Executes the program defined by an action sequence directly, without needing the overhead of translating to a logical form first. For any given program, :func:`execute` and this function are equivalent, they just take different representations of the program, so you can use whichever is more ef...
def execute_action_sequence(self, action_sequence: List[str], side_arguments: List[Dict] = None): """ Executes the program defined by an action sequence directly, without needing the overhead of translating to a logical form first. For any given program, :func:`execute` and this functio...
Induces a grammar from the defined collection of predicates in this language and returns all productions in that grammar, keyed by the non-terminal they are expanding. This includes terminal productions implied by each predicate as well as productions for the `return type` of each defined predi...
def get_nonterminal_productions(self) -> Dict[str, List[str]]: """ Induces a grammar from the defined collection of predicates in this language and returns all productions in that grammar, keyed by the non-terminal they are expanding. This includes terminal productions implied by each p...
Returns a sorted list of all production rules in the grammar induced by :func:`get_nonterminal_productions`.
def all_possible_productions(self) -> List[str]: """ Returns a sorted list of all production rules in the grammar induced by :func:`get_nonterminal_productions`. """ all_actions = set() for action_set in self.get_nonterminal_productions().values(): all_actions...
Converts a logical form into a linearization of the production rules from its abstract syntax tree. The linearization is top-down, depth-first. Each production rule is formatted as "LHS -> RHS", where "LHS" is a single non-terminal type, and RHS is either a terminal or a list of non-terminals ...
def logical_form_to_action_sequence(self, logical_form: str) -> List[str]: """ Converts a logical form into a linearization of the production rules from its abstract syntax tree. The linearization is top-down, depth-first. Each production rule is formatted as "LHS -> RHS", where "LHS" ...
Takes an action sequence as produced by :func:`logical_form_to_action_sequence`, which is a linearization of an abstract syntax tree, and reconstructs the logical form defined by that abstract syntax tree.
def action_sequence_to_logical_form(self, action_sequence: List[str]) -> str: """ Takes an action sequence as produced by :func:`logical_form_to_action_sequence`, which is a linearization of an abstract syntax tree, and reconstructs the logical form defined by that abstract syntax tree. ...
Adds a predicate to this domain language. Typically you do this with the ``@predicate`` decorator on the methods in your class. But, if you need to for whatever reason, you can also call this function yourself with a (type-annotated) function to add it to your language. Parameters ...
def add_predicate(self, name: str, function: Callable, side_arguments: List[str] = None): """ Adds a predicate to this domain language. Typically you do this with the ``@predicate`` decorator on the methods in your class. But, if you need to for whatever reason, you can also call this ...
Adds a constant to this domain language. You would typically just pass in a list of constants to the ``super().__init__()`` call in your constructor, but you can also call this method to add constants if it is more convenient. Because we construct a grammar over this language for you, in order...
def add_constant(self, name: str, value: Any, type_: Type = None): """ Adds a constant to this domain language. You would typically just pass in a list of constants to the ``super().__init__()`` call in your constructor, but you can also call this method to add constants if it is more c...
Determines whether an input symbol is a valid non-terminal in the grammar.
def is_nonterminal(self, symbol: str) -> bool: """ Determines whether an input symbol is a valid non-terminal in the grammar. """ nonterminal_productions = self.get_nonterminal_productions() return symbol in nonterminal_productions
This does the bulk of the work of executing a logical form, recursively executing a single expression. Basically, if the expression is a function we know about, we evaluate its arguments then call the function. If it's a list, we evaluate all elements of the list. If it's a constant (or a zero...
def _execute_expression(self, expression: Any): """ This does the bulk of the work of executing a logical form, recursively executing a single expression. Basically, if the expression is a function we know about, we evaluate its arguments then call the function. If it's a list, we eval...
This does the bulk of the work of :func:`execute_action_sequence`, recursively executing the functions it finds and trimming actions off of the action sequence. The return value is a tuple of (execution, remaining_actions), where the second value is necessary to handle the recursion.
def _execute_sequence(self, action_sequence: List[str], side_arguments: List[Dict]) -> Tuple[Any, List[str], List[Dict]]: """ This does the bulk of the work of :func:`execute_action_sequence`, recursively executing the functions it finds and tr...
This is used when converting a logical form into an action sequence. This piece recursively translates a lisp expression into an action sequence, making sure we match the expected type (or using the expected type to get the right type for constant expressions).
def _get_transitions(self, expression: Any, expected_type: PredicateType) -> Tuple[List[str], PredicateType]: """ This is used when converting a logical form into an action sequence. This piece recursively translates a lisp expression into an action sequence, making sure we match the ex...
A helper method for ``_get_transitions``. This gets the transitions for the predicate itself in a function call. If we only had simple functions (e.g., "(add 2 3)"), this would be pretty straightforward and we wouldn't need a separate method to handle it. We split it out into its own method b...
def _get_function_transitions(self, expression: Union[str, List], expected_type: PredicateType) -> Tuple[List[str], PredicateType, ...
Given a current node in the logical form tree, and a list of actions in an action sequence, this method fills in the children of the current node from the action sequence, then returns whatever actions are left. For example, we could get a node with type ``c``, and an action sequence that begin...
def _construct_node_from_actions(self, current_node: Tree, remaining_actions: List[List[str]]) -> List[List[str]]: """ Given a current node in the logical form tree, and a list of actions in an action sequence, this method...
Chooses ``num_samples`` samples without replacement from [0, ..., num_words). Returns a tuple (samples, num_tries).
def _choice(num_words: int, num_samples: int) -> Tuple[np.ndarray, int]: """ Chooses ``num_samples`` samples without replacement from [0, ..., num_words). Returns a tuple (samples, num_tries). """ num_tries = 0 num_chosen = 0 def get_buffer() -> np.ndarray: log_samples = np.random.r...
Takes a list of tokens and converts them to one or more sets of indices. This could be just an ID for each token from the vocabulary. Or it could split each token into characters and return one ID per character. Or (for instance, in the case of byte-pair encoding) there might not be a clean ...
def tokens_to_indices(self, tokens: List[Token], vocabulary: Vocabulary, index_name: str) -> Dict[str, List[TokenType]]: """ Takes a list of tokens and converts them to one or more sets of indices. This could be just a...
This method pads a list of tokens to ``desired_num_tokens`` and returns a padded copy of the input tokens. If the input token list is longer than ``desired_num_tokens`` then it will be truncated. ``padding_lengths`` is used to provide supplemental padding parameters which are needed in...
def pad_token_sequence(self, tokens: Dict[str, List[TokenType]], desired_num_tokens: Dict[str, int], padding_lengths: Dict[str, int]) -> Dict[str, List[TokenType]]: """ This method pads a list of tokens to ``desired_num_tok...
The CONLL 2012 data includes 2 annotated spans which are identical, but have different ids. This checks all clusters for spans which are identical, and if it finds any, merges the clusters containing the identical spans.
def canonicalize_clusters(clusters: DefaultDict[int, List[Tuple[int, int]]]) -> List[List[Tuple[int, int]]]: """ The CONLL 2012 data includes 2 annotated spans which are identical, but have different ids. This checks all clusters for spans which are identical, and if it finds any, merges the clusters co...
Join multi-word predicates to a single predicate ('V') token.
def join_mwp(tags: List[str]) -> List[str]: """ Join multi-word predicates to a single predicate ('V') token. """ ret = [] verb_flag = False for tag in tags: if "V" in tag: # Create a continuous 'V' BIO span prefix, _ = tag.split("-") if verb_flag:...
Converts a list of model outputs (i.e., a list of lists of bio tags, each pertaining to a single word), returns an inline bracket representation of the prediction.
def make_oie_string(tokens: List[Token], tags: List[str]) -> str: """ Converts a list of model outputs (i.e., a list of lists of bio tags, each pertaining to a single word), returns an inline bracket representation of the prediction. """ frame = [] chunk = [] words = [token.text for toke...
Return the word indices of a predicate in BIO tags.
def get_predicate_indices(tags: List[str]) -> List[int]: """ Return the word indices of a predicate in BIO tags. """ return [ind for ind, tag in enumerate(tags) if 'V' in tag]
Get the predicate in this prediction.
def get_predicate_text(sent_tokens: List[Token], tags: List[str]) -> str: """ Get the predicate in this prediction. """ return " ".join([sent_tokens[pred_id].text for pred_id in get_predicate_indices(tags)])
Tests whether the predicate in BIO tags1 overlap with those of tags2.
def predicates_overlap(tags1: List[str], tags2: List[str]) -> bool: """ Tests whether the predicate in BIO tags1 overlap with those of tags2. """ # Get predicate word indices from both predictions pred_ind1 = get_predicate_indices(tags1) pred_ind2 = get_predicate_indices(tags2) # Return...
Generate a coherent tag, given previous tag and current label.
def get_coherent_next_tag(prev_label: str, cur_label: str) -> str: """ Generate a coherent tag, given previous tag and current label. """ if cur_label == "O": # Don't need to add prefix to an "O" label return "O" if prev_label == cur_label: return f"I-{cur_label}" else: ...
Merge two predictions into one. Assumes the predicate in tags1 overlap with the predicate of tags2.
def merge_overlapping_predictions(tags1: List[str], tags2: List[str]) -> List[str]: """ Merge two predictions into one. Assumes the predicate in tags1 overlap with the predicate of tags2. """ ret_sequence = [] prev_label = "O" # Build a coherent sequence out of two # spans which predica...
Identify that certain predicates are part of a multiword predicate (e.g., "decided to run") in which case, we don't need to return the embedded predicate ("run").
def consolidate_predictions(outputs: List[List[str]], sent_tokens: List[Token]) -> Dict[str, List[str]]: """ Identify that certain predicates are part of a multiword predicate (e.g., "decided to run") in which case, we don't need to return the embedded predicate ("run"). """ pred_dict: Dict[str,...
Sanitize a BIO label - this deals with OIE labels sometimes having some noise, as parentheses.
def sanitize_label(label: str) -> str: """ Sanitize a BIO label - this deals with OIE labels sometimes having some noise, as parentheses. """ if "-" in label: prefix, suffix = label.split("-") suffix = suffix.split("(")[-1] return f"{prefix}-{suffix}" else: return...
Converts a batch of tokenized sentences to a tensor representing the sentences with encoded characters (len(batch), max sentence length, max word length). Parameters ---------- batch : ``List[List[str]]``, required A list of tokenized sentences. Returns ------- A tensor of padd...
def batch_to_ids(batch: List[List[str]]) -> torch.Tensor: """ Converts a batch of tokenized sentences to a tensor representing the sentences with encoded characters (len(batch), max sentence length, max word length). Parameters ---------- batch : ``List[List[str]]``, required A list of ...
Parameters ---------- inputs: ``torch.Tensor``, required. Shape ``(batch_size, timesteps, 50)`` of character ids representing the current batch. word_inputs : ``torch.Tensor``, required. If you passed a cached vocab, you can in addition pass a tensor of shape ``(b...
def forward(self, # pylint: disable=arguments-differ inputs: torch.Tensor, word_inputs: torch.Tensor = None) -> Dict[str, Union[torch.Tensor, List[torch.Tensor]]]: """ Parameters ---------- inputs: ``torch.Tensor``, required. Shape ``(batch_size...
Compute context insensitive token embeddings for ELMo representations. Parameters ---------- inputs: ``torch.Tensor`` Shape ``(batch_size, sequence_length, 50)`` of character ids representing the current batch. Returns ------- Dict with keys: ...
def forward(self, inputs: torch.Tensor) -> Dict[str, torch.Tensor]: # pylint: disable=arguments-differ """ Compute context insensitive token embeddings for ELMo representations. Parameters ---------- inputs: ``torch.Tensor`` Shape ``(batch_size, sequence_length, 50)...
Parameters ---------- inputs: ``torch.Tensor``, required. Shape ``(batch_size, timesteps, 50)`` of character ids representing the current batch. word_inputs : ``torch.Tensor``, required. If you passed a cached vocab, you can in addition pass a tensor of shape ``(batch_siz...
def forward(self, # pylint: disable=arguments-differ inputs: torch.Tensor, word_inputs: torch.Tensor = None) -> Dict[str, Union[torch.Tensor, List[torch.Tensor]]]: """ Parameters ---------- inputs: ``torch.Tensor``, required. Shape ``(batch_si...
Given a list of tokens, this method precomputes word representations by running just the character convolutions and highway layers of elmo, essentially creating uncontextual word vectors. On subsequent forward passes, the word ids are looked up from an embedding, rather than being computed on ...
def create_cached_cnn_embeddings(self, tokens: List[str]) -> None: """ Given a list of tokens, this method precomputes word representations by running just the character convolutions and highway layers of elmo, essentially creating uncontextual word vectors. On subsequent forward passes,...
Performs a normalization that is very similar to that done by the normalization functions in SQuAD and TriviaQA. This involves splitting and rejoining the text, and could be a somewhat expensive operation.
def normalize_text(text: str) -> str: """ Performs a normalization that is very similar to that done by the normalization functions in SQuAD and TriviaQA. This involves splitting and rejoining the text, and could be a somewhat expensive operation. """ return ' '.join([token ...
Converts a character span from a passage into the corresponding token span in the tokenized version of the passage. If you pass in a character span that does not correspond to complete tokens in the tokenized version, we'll do our best, but the behavior is officially undefined. We return an error flag in t...
def char_span_to_token_span(token_offsets: List[Tuple[int, int]], character_span: Tuple[int, int]) -> Tuple[Tuple[int, int], bool]: """ Converts a character span from a passage into the corresponding token span in the tokenized version of the passage. If you pass in a character ...
Finds a list of token spans in ``passage_tokens`` that match the given ``answer_texts``. This tries to find all spans that would evaluate to correct given the SQuAD and TriviaQA official evaluation scripts, which do some normalization of the input text. Note that this could return duplicate spans! The ca...
def find_valid_answer_spans(passage_tokens: List[Token], answer_texts: List[str]) -> List[Tuple[int, int]]: """ Finds a list of token spans in ``passage_tokens`` that match the given ``answer_texts``. This tries to find all spans that would evaluate to correct given the SQuAD an...
Converts a question, a passage, and an optional answer (or answers) to an ``Instance`` for use in a reading comprehension model. Creates an ``Instance`` with at least these fields: ``question`` and ``passage``, both ``TextFields``; and ``metadata``, a ``MetadataField``. Additionally, if both ``answer_text...
def make_reading_comprehension_instance(question_tokens: List[Token], passage_tokens: List[Token], token_indexers: Dict[str, TokenIndexer], passage_text: str, t...
Converts a question, a passage, and an optional answer (or answers) to an ``Instance`` for use in a reading comprehension model. Creates an ``Instance`` with at least these fields: ``question`` and ``passage``, both ``TextFields``; and ``metadata``, a ``MetadataField``. Additionally, if both ``answer_text...
def make_reading_comprehension_instance_quac(question_list_tokens: List[List[Token]], passage_tokens: List[Token], token_indexers: Dict[str, TokenIndexer], passage_text: str, ...
Process a list of reference answers. If equal or more than half of the reference answers are "CANNOTANSWER", take it as gold. Otherwise, return answers that are not "CANNOTANSWER".
def handle_cannot(reference_answers: List[str]): """ Process a list of reference answers. If equal or more than half of the reference answers are "CANNOTANSWER", take it as gold. Otherwise, return answers that are not "CANNOTANSWER". """ num_cannot = 0 num_spans = 0 for ref in reference_...
This acts the same as the static method ``BidirectionalAttentionFlow.get_best_span()`` in ``allennlp/models/reading_comprehension/bidaf.py``. We keep it here so that users can directly import this function without the class. We call the inputs "logits" - they could either be unnormalized logits or normaliz...
def get_best_span(span_start_logits: torch.Tensor, span_end_logits: torch.Tensor) -> torch.Tensor: """ This acts the same as the static method ``BidirectionalAttentionFlow.get_best_span()`` in ``allennlp/models/reading_comprehension/bidaf.py``. We keep it here so that users can directly import this func...
Spacy needs to do batch processing, or it can be really slow. This method lets you take advantage of that if you want. Default implementation is to just iterate of the sentences and call ``split_words``, but the ``SpacyWordSplitter`` will actually do batched processing.
def batch_split_words(self, sentences: List[str]) -> List[List[Token]]: """ Spacy needs to do batch processing, or it can be really slow. This method lets you take advantage of that if you want. Default implementation is to just iterate of the sentences and call ``split_words``, but th...
Return a new BeamSearch instance that's like this one but with the specified constraint.
def constrained_to(self, initial_sequence: torch.Tensor, keep_beam_details: bool = True) -> 'BeamSearch': """ Return a new BeamSearch instance that's like this one but with the specified constraint. """ return BeamSearch(self._beam_size, self._per_node_beam_size, initial_sequence, keep_b...
Parameters ---------- num_steps : ``int`` How many steps should we take in our search? This is an upper bound, as it's possible for the search to run out of valid actions before hitting this number, or for all states on the beam to finish. initial_state : ``S...
def search(self, num_steps: int, initial_state: StateType, transition_function: TransitionFunction, keep_final_unfinished_states: bool = True) -> Dict[int, List[StateType]]: """ Parameters ---------- num_steps : ``int`` ...
Lower text and remove punctuation, articles and extra whitespace.
def _normalize_answer(text: str) -> str: """Lower text and remove punctuation, articles and extra whitespace.""" parts = [_white_space_fix(_remove_articles(_normalize_number(_remove_punc(_lower(token))))) for token in _tokenize(text)] parts = [part for part in parts if part.strip()] normal...
Takes gold and predicted answer sets and first finds a greedy 1-1 alignment between them and gets maximum metric values over all the answers
def _align_bags(predicted: List[Set[str]], gold: List[Set[str]]) -> List[float]: """ Takes gold and predicted answer sets and first finds a greedy 1-1 alignment between them and gets maximum metric values over all the answers """ f1_scores = [] for gold_index, gold_item in enumerate(gold): ...
Takes a predicted answer and a gold answer (that are both either a string or a list of strings), and returns exact match and the DROP F1 metric for the prediction. If you are writing a script for evaluating objects in memory (say, the output of predictions during validation, or while training), this is the...
def get_metrics(predicted: Union[str, List[str], Tuple[str, ...]], gold: Union[str, List[str], Tuple[str, ...]]) -> Tuple[float, float]: """ Takes a predicted answer and a gold answer (that are both either a string or a list of strings), and returns exact match and the DROP F1 metric for the...
Takes an answer JSON blob from the DROP data release and converts it into strings used for evaluation.
def answer_json_to_strings(answer: Dict[str, Any]) -> Tuple[Tuple[str, ...], str]: """ Takes an answer JSON blob from the DROP data release and converts it into strings used for evaluation. """ if "number" in answer and answer["number"]: return tuple([str(answer["number"])]), "number" el...
Takes a prediction file and a gold file and evaluates the predictions for each question in the gold file. Both files must be json formatted and must have query_id keys, which are used to match predictions to gold annotations. The gold file is assumed to have the format of the dev set in the DROP data rele...
def evaluate_prediction_file(prediction_path: str, gold_path: str) -> Tuple[float, float]: """ Takes a prediction file and a gold file and evaluates the predictions for each question in the gold file. Both files must be json formatted and must have query_id keys, which are used to match predictions to ...
When you call this method, we will use this directory to store a cache of already-processed ``Instances`` in every file passed to :func:`read`, serialized as one string-formatted ``Instance`` per line. If the cache file for a given ``file_path`` exists, we read the ``Instances`` from the cache ...
def cache_data(self, cache_directory: str) -> None: """ When you call this method, we will use this directory to store a cache of already-processed ``Instances`` in every file passed to :func:`read`, serialized as one string-formatted ``Instance`` per line. If the cache file for a given...
Returns an ``Iterable`` containing all the instances in the specified dataset. If ``self.lazy`` is False, this calls ``self._read()``, ensures that the result is a list, then returns the resulting list. If ``self.lazy`` is True, this returns an object whose ``__iter__`` method ...
def read(self, file_path: str) -> Iterable[Instance]: """ Returns an ``Iterable`` containing all the instances in the specified dataset. If ``self.lazy`` is False, this calls ``self._read()``, ensures that the result is a list, then returns the resulting list. If ``self...
Restores a model from a serialization_dir to the last saved checkpoint. This includes a training state (typically consisting of an epoch count and optimizer state), which is serialized separately from model parameters. This function should only be used to continue training - if you wish to load...
def restore_checkpoint(self) -> Tuple[Dict[str, Any], Dict[str, Any]]: """ Restores a model from a serialization_dir to the last saved checkpoint. This includes a training state (typically consisting of an epoch count and optimizer state), which is serialized separately from model param...
Prints predicate argument predictions and gold labels for a single verbal predicate in a sentence to two provided file references. Parameters ---------- prediction_file : TextIO, required. A file reference to print predictions to. gold_file : TextIO, required. A file reference to pr...
def write_to_conll_eval_file(prediction_file: TextIO, gold_file: TextIO, verb_index: Optional[int], sentence: List[str], prediction: List[str], gold_labels: List[str]): ""...
Converts BIO formatted SRL tags to the format required for evaluation with the official CONLL 2005 perl script. Spans are represented by bracketed labels, with the labels of words inside spans being the same as those outside spans. Beginning spans always have a opening bracket and a closing asterisk (e.g. "...
def convert_bio_tags_to_conll_format(labels: List[str]): """ Converts BIO formatted SRL tags to the format required for evaluation with the official CONLL 2005 perl script. Spans are represented by bracketed labels, with the labels of words inside spans being the same as those outside spans. Beginni...
Given a ``sentence``, returns a list of actions the sentence triggers as an ``agenda``. The ``agenda`` can be used while by a parser to guide the decoder. sequences as possible. This is a simplistic mapping at this point, and can be expanded. Parameters ---------- sentence : ``...
def get_agenda_for_sentence(self, sentence: str) -> List[str]: """ Given a ``sentence``, returns a list of actions the sentence triggers as an ``agenda``. The ``agenda`` can be used while by a parser to guide the decoder. sequences as possible. This is a simplistic mapping at this point...
Gathers all the numbers in the sentence, and returns productions that lead to them.
def _get_number_productions(sentence: str) -> List[str]: """ Gathers all the numbers in the sentence, and returns productions that lead to them. """ # The mapping here is very simple and limited, which also shouldn't be a problem # because numbers seem to be represented fairly re...
Filters the set of objects, and returns those objects whose color is the most frequent color in the initial set of objects, if the highest frequency is greater than 1, or an empty set otherwise. This is an unusual name for what the method does, but just as ``blue`` filters objects to th...
def same_color(self, objects: Set[Object]) -> Set[Object]: """ Filters the set of objects, and returns those objects whose color is the most frequent color in the initial set of objects, if the highest frequency is greater than 1, or an empty set otherwise. This is an unusual na...
Filters the set of objects, and returns those objects whose color is the most frequent color in the initial set of objects, if the highest frequency is greater than 1, or an empty set otherwise. This is an unusual name for what the method does, but just as ``triangle`` filters objects t...
def same_shape(self, objects: Set[Object]) -> Set[Object]: """ Filters the set of objects, and returns those objects whose color is the most frequent color in the initial set of objects, if the highest frequency is greater than 1, or an empty set otherwise. This is an unusual na...
Returns all objects that touch the given set of objects.
def touch_object(self, objects: Set[Object]) -> Set[Object]: """ Returns all objects that touch the given set of objects. """ objects_per_box = self._separate_objects_by_boxes(objects) return_set = set() for box, box_objects in objects_per_box.items(): candida...
Return the topmost objects (i.e. minimum y_loc). The comparison is done separately for each box.
def top(self, objects: Set[Object]) -> Set[Object]: """ Return the topmost objects (i.e. minimum y_loc). The comparison is done separately for each box. """ objects_per_box = self._separate_objects_by_boxes(objects) return_set: Set[Object] = set() for _, box_objec...
Return the bottom most objects(i.e. maximum y_loc). The comparison is done separately for each box.
def bottom(self, objects: Set[Object]) -> Set[Object]: """ Return the bottom most objects(i.e. maximum y_loc). The comparison is done separately for each box. """ objects_per_box = self._separate_objects_by_boxes(objects) return_set: Set[Object] = set() for _, box...
Returns the set of objects in the same boxes that are above the given objects. That is, if the input is a set of two objects, one in each box, we will return a union of the objects above the first object in the first box, and those above the second object in the second box.
def above(self, objects: Set[Object]) -> Set[Object]: """ Returns the set of objects in the same boxes that are above the given objects. That is, if the input is a set of two objects, one in each box, we will return a union of the objects above the first object in the first box, and thos...
Returns the set of objects in the same boxes that are below the given objects. That is, if the input is a set of two objects, one in each box, we will return a union of the objects below the first object in the first box, and those below the second object in the second box.
def below(self, objects: Set[Object]) -> Set[Object]: """ Returns the set of objects in the same boxes that are below the given objects. That is, if the input is a set of two objects, one in each box, we will return a union of the objects below the first object in the first box, and thos...
Returns true iff the objects touch each other.
def _objects_touch_each_other(self, object1: Object, object2: Object) -> bool: """ Returns true iff the objects touch each other. """ in_vertical_range = object1.y_loc <= object2.y_loc + object2.size and \ object1.y_loc + object1.size >= object2.y_loc ...
Given a set of objects, separate them by the boxes they belong to and return a dict.
def _separate_objects_by_boxes(self, objects: Set[Object]) -> Dict[Box, List[Object]]: """ Given a set of objects, separate them by the boxes they belong to and return a dict. """ objects_per_box: Dict[Box, List[Object]] = defaultdict(list) for box in self.boxes: for ...
Returns the set of objects for which the attribute function returns an attribute value that is most frequent in the initial set, if the frequency is greater than 1. If not, all objects have different attribute values, and this method returns an empty set.
def _get_objects_with_same_attribute(self, objects: Set[Object], attribute_function: Callable[[Object], str]) -> Set[Object]: """ Returns the set of objects for which the attribute function returns an attribute value that ...
Given a structure (possibly) containing Tensors on the CPU, move all the Tensors to the specified GPU (or do nothing, if they should be on the CPU).
def move_to_device(obj, cuda_device: int): """ Given a structure (possibly) containing Tensors on the CPU, move all the Tensors to the specified GPU (or do nothing, if they should be on the CPU). """ if cuda_device < 0 or not has_tensor(obj): return obj elif isinstance(obj, torch.Tensor)...
Given a possibly complex data structure, check if it has any torch.Tensors in it.
def has_tensor(obj) -> bool: """ Given a possibly complex data structure, check if it has any torch.Tensors in it. """ if isinstance(obj, torch.Tensor): return True elif isinstance(obj, dict): return any(has_tensor(value) for value in obj.values()) elif isinstance(obj, (list,...
Supports sparse and dense tensors. Returns a tensor with values clamped between the provided minimum and maximum, without modifying the original tensor.
def clamp_tensor(tensor, minimum, maximum): """ Supports sparse and dense tensors. Returns a tensor with values clamped between the provided minimum and maximum, without modifying the original tensor. """ if tensor.is_sparse: coalesced_tensor = tensor.coalesce() # pylint: disable...
Takes a list of tensor dictionaries, where each dictionary is assumed to have matching keys, and returns a single dictionary with all tensors with the same key batched together. Parameters ---------- tensor_dicts : ``List[Dict[str, torch.Tensor]]`` The list of tensor dictionaries to batch. ...
def batch_tensor_dicts(tensor_dicts: List[Dict[str, torch.Tensor]], remove_trailing_dimension: bool = False) -> Dict[str, torch.Tensor]: """ Takes a list of tensor dictionaries, where each dictionary is assumed to have matching keys, and returns a single dictionary with all tensors wi...
Given a variable of shape ``(batch_size,)`` that represents the sequence lengths of each batch element, this function returns a ``(batch_size, max_length)`` mask variable. For example, if our input was ``[2, 2, 3]``, with a ``max_length`` of 4, we'd return ``[[1, 1, 0, 0], [1, 1, 0, 0], [1, 1, 1, 0]]``. ...
def get_mask_from_sequence_lengths(sequence_lengths: torch.Tensor, max_length: int) -> torch.Tensor: """ Given a variable of shape ``(batch_size,)`` that represents the sequence lengths of each batch element, this function returns a ``(batch_size, max_length)`` mask variable. For example, if our input ...