Pipeline

SpanCategorizer

class,experimentalv3.1
String name:spancatTrainable:
Pipeline component for labeling potentially overlapping spans of text

A span categorizer consists of two parts: a suggester function that proposes candidate spans, which may or may not overlap, and a labeler model that predicts zero or more labels for each candidate.

This component comes in two forms: spancat and spancat_singlelabel (added in spaCy v3.5.1). When you need to perform multi-label classification on your spans, use spancat. The spancat component uses a Logistic layer where the output class probabilities are independent for each class. However, if you need to predict at most one true class for a span, then use spancat_singlelabel. It uses a Softmax layer and treats the task as a multi-class problem.

Predicted spans will be saved in a SpanGroup on the doc under doc.spans[spans_key], where spans_key is a component config setting. Individual span scores are stored in doc.spans[spans_key].attrs["scores"].

Assigned Attributes

Predictions will be saved to Doc.spans[spans_key] as a SpanGroup. The scores for the spans in the SpanGroup will be saved in SpanGroup.attrs["scores"].

spans_key defaults to "sc", but can be passed as a parameter. The spancat component will overwrite any existing spans under the spans key doc.spans[spans_key].

LocationValue
Doc.spans[spans_key]The annotated spans. SpanGroup
Doc.spans[spans_key].attrs["scores"]The score for each span in the SpanGroup. Floats1d

Config and implementation

The default config is defined by the pipeline component factory and describes how the component should be configured. You can override its settings via the config argument on nlp.add_pipe or in your config.cfg for training. See the model architectures documentation for details on the architectures and their arguments and hyperparameters.

SettingDescription
suggesterA function that suggests spans. Spans are returned as a ragged array with two integer columns, for the start and end positions. Defaults to ngram_suggester. Callable[[Iterable[Doc], Optional[Ops]],Ragged]
modelA model instance that is given a a list of documents and (start, end) indices representing candidate span offsets. The model predicts a probability for each category for each span. Defaults to SpanCategorizer. Model[Tuple[List[Doc],Ragged],Floats2d]
spans_keyKey of the Doc.spans dict to save the spans under. During initialization and training, the component will look for spans on the reference document under the same key. Defaults to "sc". str
thresholdMinimum probability to consider a prediction positive. Spans with a positive prediction will be saved on the Doc. Meant to be used in combination with the multi-class spancat component with a Logistic scoring layer. Defaults to 0.5. float
max_positiveMaximum number of labels to consider positive per span. Defaults to None, indicating no limit. Meant to be used together with the spancat component and defaults to 0 with spancat_singlelabel. Optional[int]
scorerThe scoring method. Defaults to Scorer.score_spans for Doc.spans[spans_key] with overlapping spans allowed. Optional[Callable]
add_negative_label v3.5.1Whether to learn to predict a special negative label for each unannotated Span . This should be True when using a Softmax classifier layer and so its True by default for spancat_singlelabel. Spans with negative labels and their scores are not stored as annotations. bool
negative_weight v3.5.1Multiplier for the loss terms. It can be used to downweight the negative samples if there are too many. It is only used when add_negative_label is True. Defaults to 1.0. float
allow_overlap v3.5.1If True, the data is assumed to contain overlapping spans. It is only available when max_positive is exactly 1. Defaults to True. bool
explosion/spaCy/master/spacy/pipeline/spancat.py

SpanCategorizer.__init__ method

Create a new pipeline instance. In your application, you would normally use a shortcut for this and instantiate the component using its string name and nlp.add_pipe.

NameDescription
vocabThe shared vocabulary. Vocab
modelA model instance that is given a a list of documents and (start, end) indices representing candidate span offsets. The model predicts a probability for each category for each span. Model[Tuple[List[Doc],Ragged],Floats2d]
suggesterA function that suggests spans. Spans are returned as a ragged array with two integer columns, for the start and end positions. Callable[[Iterable[Doc], Optional[Ops]],Ragged]
nameString name of the component instance. Used to add entries to the losses during training. str
keyword-only
spans_keyKey of the Doc.spans dict to save the spans under. During initialization and training, the component will look for spans on the reference document under the same key. Defaults to "sc". str
thresholdMinimum probability to consider a prediction positive. Spans with a positive prediction will be saved on the Doc. Defaults to 0.5. float
max_positiveMaximum number of labels to consider positive per span. Defaults to None, indicating no limit. Optional[int]
allow_overlap v3.5.1If True, the data is assumed to contain overlapping spans. It is only available when max_positive is exactly 1. Defaults to True. bool
add_negative_label v3.5.1Whether to learn to predict a special negative label for each unannotated Span. This should be True when using a Softmax classifier layer and so its True by default for spancat_singlelabel . Spans with negative labels and their scores are not stored as annotations. bool
negative_weight v3.5.1Multiplier for the loss terms. It can be used to downweight the negative samples if there are too many . It is only used when add_negative_label is True. Defaults to 1.0. float

SpanCategorizer.__call__ method

Apply the pipe to one document. The document is modified in place, and returned. This usually happens under the hood when the nlp object is called on a text and all pipeline components are applied to the Doc in order. Both __call__ and pipe delegate to the predict and set_annotations methods.

NameDescription
docThe document to process. Doc

SpanCategorizer.pipe method

Apply the pipe to a stream of documents. This usually happens under the hood when the nlp object is called on a text and all pipeline components are applied to the Doc in order. Both __call__ and pipe delegate to the predict and set_annotations methods.

NameDescription
streamA stream of documents. Iterable[Doc]
keyword-only
batch_sizeThe number of documents to buffer. Defaults to 128. int

SpanCategorizer.initialize method

Initialize the component for training. get_examples should be a function that returns an iterable of Example objects. At least one example should be supplied. The data examples are used to initialize the model of the component and can either be the full training data or a representative sample. Initialization includes validating the network, inferring missing shapes and setting up the label scheme based on the data. This method is typically called by Language.initialize and lets you customize arguments it receives via the [initialize.components] block in the config.

NameDescription
get_examplesFunction that returns gold-standard annotations in the form of Example objects. Must contain at least one Example. Callable[[], Iterable[Example]]
keyword-only
nlpThe current nlp object. Defaults to None. Optional[Language]
labelsThe label information to add to the component, as provided by the label_data property after initialization. To generate a reusable JSON file from your data, you should run the init labels command. If no labels are provided, the get_examples callback is used to extract the labels from the data, which may be a lot slower. Optional[Iterable[str]]

SpanCategorizer.predict method

Apply the component’s model to a batch of Doc objects without modifying them.

NameDescription
docsThe documents to predict. Iterable[Doc]

SpanCategorizer.set_annotations method

Modify a batch of Doc objects using pre-computed scores.

NameDescription
docsThe documents to modify. Iterable[Doc]
scoresThe scores to set, produced by SpanCategorizer.predict.

SpanCategorizer.update method

Learn from a batch of Example objects containing the predictions and gold-standard annotations, and update the component’s model. Delegates to predict and get_loss.

NameDescription
examplesA batch of Example objects to learn from. Iterable[Example]
keyword-only
dropThe dropout rate. float
sgdAn optimizer. Will be created via create_optimizer if not set. Optional[Optimizer]
lossesOptional record of the loss during training. Updated using the component name as the key. Optional[Dict[str, float]]

SpanCategorizer.set_candidates methodv3.3

Use the suggester to add a list of Span candidates to a list of Doc objects. This method is intended to be used for debugging purposes.

NameDescription
docsThe documents to modify. Iterable[Doc]
candidates_keyKey of the Doc.spans dict to save the candidate spans under. str

SpanCategorizer.get_loss method

Find the loss and gradient of loss for the batch of documents and their predicted scores.

NameDescription
examplesThe batch of examples. Iterable[Example]
spans_scoresScores representing the model’s predictions. Tuple[Ragged,Floats2d]

SpanCategorizer.create_optimizer method

Create an optimizer for the pipeline component.

NameDescription

SpanCategorizer.use_params methodcontextmanager

Modify the pipe’s model to use the given parameter values.

NameDescription
paramsThe parameter values to use in the model. dict

SpanCategorizer.add_label method

Add a new label to the pipe. Raises an error if the output dimension is already set, or if the model has already been fully initialized. Note that you don’t have to call this method if you provide a representative data sample to the initialize method. In this case, all labels found in the sample will be automatically added to the model, and the output dimension will be inferred automatically.

NameDescription
labelThe label to add. str

SpanCategorizer.to_disk method

Serialize the pipe to disk.

NameDescription
pathA path to a directory, which will be created if it doesn’t exist. Paths may be either strings or Path-like objects. Union[str,Path]
keyword-only
excludeString names of serialization fields to exclude. Iterable[str]

SpanCategorizer.from_disk method

Load the pipe from disk. Modifies the object in place and returns it.

NameDescription
pathA path to a directory. Paths may be either strings or Path-like objects. Union[str,Path]
keyword-only
excludeString names of serialization fields to exclude. Iterable[str]

SpanCategorizer.to_bytes method

Serialize the pipe to a bytestring.

NameDescription
keyword-only
excludeString names of serialization fields to exclude. Iterable[str]

SpanCategorizer.from_bytes method

Load the pipe from a bytestring. Modifies the object in place and returns it.

NameDescription
bytes_dataThe data to load from. bytes
keyword-only
excludeString names of serialization fields to exclude. Iterable[str]

SpanCategorizer.labels property

The labels currently added to the component.

NameDescription

SpanCategorizer.label_data property

The labels currently added to the component and their internal meta information. This is the data generated by init labels and used by SpanCategorizer.initialize to initialize the model with a pre-defined label set.

NameDescription

Serialization fields

During serialization, spaCy will export several data fields used to restore different aspects of the object. If needed, you can exclude them from serialization by passing in the string names via the exclude argument.

NameDescription
vocabThe shared Vocab.
cfgThe config file. You usually don’t want to exclude this.
modelThe binary model data. You usually don’t want to exclude this.

Suggesters registered functions

spacy.ngram_suggester.v1

Suggest all spans of the given lengths. Spans are returned as a ragged array of integers. The array has two columns, indicating the start and end position.

NameDescription
sizesThe phrase lengths to suggest. For example, [1, 2] will suggest phrases consisting of 1 or 2 tokens. List[int]

spacy.ngram_range_suggester.v1

Suggest all spans of at least length min_size and at most length max_size (both inclusive). Spans are returned as a ragged array of integers. The array has two columns, indicating the start and end position.

NameDescription
min_sizeThe minimal phrase lengths to suggest (inclusive). [int]
max_sizeThe maximal phrase lengths to suggest (inclusive). [int]

spacy.preset_spans_suggester.v1

Suggest all spans that are already stored in doc.spans[spans_key]. This is useful when an upstream component is used to set the spans on the Doc such as a SpanRuler or SpanFinder.

NameDescription
spans_keyKey of Doc.spans that provides spans to suggest. str