spaCy/website/docs/api/pipe.md

title	tag	teaser
TrainablePipe	class	Base class for trainable pipeline components

This class is a base class and not instantiated directly. Trainable pipeline components like the EntityRecognizer or TextCategorizer inherit from it and it defines the interface that components should follow to function as trainable components in a spaCy pipeline. See the docs on writing trainable components for how to use the TrainablePipe base class to implement custom components.

Why is it implemented in Cython?

The TrainablePipe class is implemented in a .pyx module, the extension used by Cython. This is needed so that other Cython classes, like the EntityRecognizer can inherit from it. But it doesn't mean you have to implement trainable components in Cython – pure Python components like the TextCategorizer can also inherit from TrainablePipe.

%%GITHUB_SPACY/spacy/pipeline/trainable_pipe.pyx

TrainablePipe.__init__

Example

from spacy.pipeline import TrainablePipe
from spacy.language import Language

class CustomPipe(TrainablePipe):
    ...

@Language.factory("your_custom_pipe", default_config={"model": MODEL})
def make_custom_pipe(nlp, name, model):
    return CustomPipe(nlp.vocab, model, name)

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.

Name	Description
vocab	The shared vocabulary. Vocab
model	The Thinc Model powering the pipeline component. Model[List[Doc], Any]
name	String name of the component instance. Used to add entries to the losses during training. str
**cfg	Additional config parameters and settings. Will be available as the dictionary cfg and is serialized with the component.

TrainablePipe.__call__

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.

Example

doc = nlp("This is a sentence.")
pipe = nlp.add_pipe("your_custom_pipe")
# This usually happens under the hood
processed = pipe(doc)

Name	Description
doc	The document to process. Doc
RETURNS	The processed document. Doc

TrainablePipe.pipe

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.

Example

pipe = nlp.add_pipe("your_custom_pipe")
for doc in pipe.pipe(docs, batch_size=50):
    pass

Name	Description
stream	A stream of documents. Iterable[Doc]
keyword-only
batch_size	The number of documents to buffer. Defaults to 128. int
YIELDS	The processed documents in order. Doc

TrainablePipe.set_error_handler

Define a callback that will be invoked when an error is thrown during processing of one or more documents with either __call__ or pipe. The error handler will be invoked with the original component's name, the component itself, the list of documents that was being processed, and the original error.

Example

def warn_error(proc_name, proc, docs, e):
    print(f"An error occurred when applying component {proc_name}.")

pipe = nlp.add_pipe("ner")
pipe.set_error_handler(warn_error)

Name	Description
error_handler	A function that performs custom error handling. Callable[[str, Callable[[Doc], Doc], List[Doc], Exception]

TrainablePipe.get_error_handler

Retrieve the callback that performs error handling for this component's __call__ and pipe methods. If no custom function was previously defined with set_error_handler, a default function is returned that simply reraises the exception.

Example

pipe = nlp.add_pipe("ner")
error_handler = pipe.get_error_handler()

Name	Description
RETURNS	The function that performs custom error handling. Callable[[str, Callable[[Doc], Doc], List[Doc], Exception]

TrainablePipe.initialize

Initialize the component for training. get_examples should be a function that returns an iterable of Example objects. 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.

This method was previously called begin_training.

Example

pipe = nlp.add_pipe("your_custom_pipe")
pipe.initialize(lambda: [], pipeline=nlp.pipeline)

Name	Description
get_examples	Function that returns gold-standard annotations in the form of Example objects. Callable], Iterable[Example
keyword-only
nlp	The current nlp object. Defaults to None. Optional[Language]

TrainablePipe.predict

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

This method needs to be overwritten with your own custom predict method.

Example

pipe = nlp.add_pipe("your_custom_pipe")
scores = pipe.predict([doc1, doc2])

Name	Description
docs	The documents to predict. Iterable[Doc]
RETURNS	The model's prediction for each document.

TrainablePipe.set_annotations

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

This method needs to be overwritten with your own custom set_annotations method.

Example

pipe = nlp.add_pipe("your_custom_pipe")
scores = pipe.predict(docs)
pipe.set_annotations(docs, scores)

Name	Description
docs	The documents to modify. Iterable[Doc]
scores	The scores to set, produced by Tagger.predict.

TrainablePipe.update

Learn from a batch of Example objects containing the predictions and gold-standard annotations, and update the component's model.

Example

pipe = nlp.add_pipe("your_custom_pipe")
optimizer = nlp.initialize()
losses = pipe.update(examples, sgd=optimizer)

Name	Description
examples	A batch of Example objects to learn from. Iterable[Example]
keyword-only
drop	The dropout rate. float
sgd	An optimizer. Will be created via create_optimizer if not set. Optional[Optimizer]
losses	Optional record of the loss during training. Updated using the component name as the key. Optional[Dict[str, float]]
RETURNS	The updated losses dictionary. Dict[str, float]

TrainablePipe.rehearse

Perform a "rehearsal" update from a batch of data. Rehearsal updates teach the current model to make predictions similar to an initial model, to try to address the "catastrophic forgetting" problem. This feature is experimental.

Example

pipe = nlp.add_pipe("your_custom_pipe")
optimizer = nlp.resume_training()
losses = pipe.rehearse(examples, sgd=optimizer)

Name	Description
examples	A batch of Example objects to learn from. Iterable[Example]
keyword-only
sgd	An optimizer. Will be created via create_optimizer if not set. Optional[Optimizer]
losses	Optional record of the loss during training. Updated using the component name as the key. Optional[Dict[str, float]]
RETURNS	The updated losses dictionary. Dict[str, float]

TrainablePipe.get_loss

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

This method needs to be overwritten with your own custom get_loss method.

Example

ner = nlp.add_pipe("ner")
scores = ner.predict([eg.predicted for eg in examples])
loss, d_loss = ner.get_loss(examples, scores)

Name	Description
examples	The batch of examples. Iterable[Example]
scores	Scores representing the model's predictions.
RETURNS	The loss and the gradient, i.e. (loss, gradient). Tuple[float, float]

TrainablePipe.score

Score a batch of examples.

Example

scores = pipe.score(examples)

Name	Description
examples	The examples to score. Iterable[Example]
keyword-only
\*\*kwargs	Any additional settings to pass on to the scorer. Any
RETURNS	The scores, e.g. produced by the Scorer. Dict[str, Union[float, Dict[str, float]]]

TrainablePipe.create_optimizer

Create an optimizer for the pipeline component. Defaults to Adam with default settings.

Example

pipe = nlp.add_pipe("your_custom_pipe")
optimizer = pipe.create_optimizer()

Name	Description
RETURNS	The optimizer. Optimizer

TrainablePipe.use_params

Modify the pipe's model, to use the given parameter values. At the end of the context, the original parameters are restored.

Example

pipe = nlp.add_pipe("your_custom_pipe")
with pipe.use_params(optimizer.averages):
    pipe.to_disk("/best_model")

Name	Description
params	The parameter values to use in the model. dict

TrainablePipe.finish_update

Update parameters using the current parameter gradients. Defaults to calling self.model.finish_update.

Example

pipe = nlp.add_pipe("your_custom_pipe")
optimizer = nlp.initialize()
losses = pipe.update(examples, sgd=None)
pipe.finish_update(sgd)

Name	Description
sgd	An optimizer. Optional[Optimizer]

TrainablePipe.add_label

Example

pipe = nlp.add_pipe("your_custom_pipe")
pipe.add_label("MY_LABEL")

Add a new label to the pipe, to be predicted by the model. The actual implementation depends on the specific component, but in general add_label shouldn't be called if the output dimension is already set, or if the model has already been fully initialized. If these conditions are violated, the function will raise an Error. The exception to this rule is when the component is resizable, in which case set_output should be called to ensure that the model is properly resized.

This method needs to be overwritten with your own custom add_label method.

Name	Description
label	The label to add. str
RETURNS	0 if the label is already present, otherwise 1. int

Note that in general, you don't have to call pipe.add_label 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.

TrainablePipe.is_resizable

Example

can_resize = pipe.is_resizable

With custom resizing implemented by a component:

def custom_resize(model, new_nO):
    # adjust model
    return model

custom_model.attrs["resize_output"] = custom_resize

Check whether or not the output dimension of the component's model can be resized. If this method returns True, set_output can be called to change the model's output dimension.

For built-in components that are not resizable, you have to create and train a new model from scratch with the appropriate architecture and output dimension. For custom components, you can implement a resize_output function and add it as an attribute to the component's model.

Name	Description
RETURNS	Whether or not the output dimension of the model can be changed after initialization. bool

TrainablePipe.set_output

Change the output dimension of the component's model. If the component is not resizable, this method will raise a NotImplementedError. If a component is resizable, the model's attribute resize_output will be called. This is a function that takes the original model and the new output dimension nO, and changes the model in place. When resizing an already trained model, care should be taken to avoid the "catastrophic forgetting" problem.

Example

if pipe.is_resizable:
    pipe.set_output(512)

Name	Description
nO	The new output dimension. int

TrainablePipe.to_disk

Serialize the pipe to disk.

Example

pipe = nlp.add_pipe("your_custom_pipe")
pipe.to_disk("/path/to/pipe")

Name	Description
path	A 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
exclude	String names of serialization fields to exclude. Iterable[str]

TrainablePipe.from_disk

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

Example

pipe = nlp.add_pipe("your_custom_pipe")
pipe.from_disk("/path/to/pipe")

Name	Description
path	A path to a directory. Paths may be either strings or Path-like objects. Union[str, Path]
keyword-only
exclude	String names of serialization fields to exclude. Iterable[str]
RETURNS	The modified pipe. TrainablePipe

TrainablePipe.to_bytes

Example

pipe = nlp.add_pipe("your_custom_pipe")
pipe_bytes = pipe.to_bytes()

Serialize the pipe to a bytestring.

Name	Description
keyword-only
exclude	String names of serialization fields to exclude. Iterable[str]
RETURNS	The serialized form of the pipe. bytes

TrainablePipe.from_bytes

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

Example

pipe_bytes = pipe.to_bytes()
pipe = nlp.add_pipe("your_custom_pipe")
pipe.from_bytes(pipe_bytes)

Name	Description
bytes_data	The data to load from. bytes
keyword-only
exclude	String names of serialization fields to exclude. Iterable[str]
RETURNS	The pipe. TrainablePipe

Attributes

Name	Description
vocab	The shared vocabulary that's passed in on initialization. Vocab
model	The model powering the component. Model[List[Doc], Any]
name	The name of the component instance in the pipeline. Can be used in the losses. str
cfg	Keyword arguments passed to TrainablePipe.__init__. Will be serialized with the component. Dict[str, Any]

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.

Example

data = pipe.to_disk("/path")

Name	Description
cfg	The config file. You usually don't want to exclude this.
model	The binary model data. You usually don't want to exclude this.