explosion/spaCy
1
1
Fork 0
mirror of https://github.com/explosion/spaCy.git synced 2025-01-13 02:36:32 +03:00
Code Issues Packages Projects Releases Wiki Activity

Merge pull request #6419 from svlandeg/feature/rel-docs

Browse Source
This commit is contained in:
Ines Montani 2020-12-08 06:30:41 +01:00 committed by GitHub
commit c2b196c2c1
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
1 changed files with 263 additions and 140 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

403
website/docs/usage/layers-architectures.md
View File

@ -502,7 +502,7 @@ with Model.define_operators({">>": chain}):
## Create new trainable components {#components} ## Create new trainable components {#components}
In addition to [swapping out](#swap-architectures) default models in built-in In addition to [swapping out](#swap-architectures) layers in existing
components, you can also implement an entirely new, components, you can also implement an entirely new,
[trainable](/usage/processing-pipelines#trainable-components) pipeline component [trainable](/usage/processing-pipelines#trainable-components) pipeline component
from scratch. This can be done by creating a new class inheriting from from scratch. This can be done by creating a new class inheriting from
@ -523,20 +523,28 @@ overview of the `TrainablePipe` methods used by
This section outlines an example use-case of implementing a **novel relation This section outlines an example use-case of implementing a **novel relation
extraction component** from scratch. We'll implement a binary relation extraction component** from scratch. We'll implement a binary relation
extraction method that determines whether or not **two entities** in a document extraction method that determines whether or not **two entities** in a document
are related, and if so, what type of relation. We'll allow multiple types of are related, and if so, what type of relation connects them. We allow multiple
relations between two such entities (multi-label setting). There are two major types of relations between two such entities (a multi-label setting). There are
steps required: two major steps required:
1. Implement a [machine learning model](#component-rel-model) specific to this 1. Implement a [machine learning model](#component-rel-model) specific to this
task. It will have to extract candidates from a [`Doc`](/api/doc) and predict task. It will have to extract candidate relation instances from a
a relation for the available candidate pairs. [`Doc`](/api/doc) and predict the corresponding scores for each relation
2. Implement a custom [pipeline component](#component-rel-pipe) powered by the label.
machine learning model that sets annotations on the [`Doc`](/api/doc) passing 2. Implement a custom [pipeline component](#component-rel-pipe) - powered by the
through the pipeline. machine learning model from step 1 - that translates the predicted scores
into annotations that are stored on the [`Doc`](/api/doc) objects as they
pass through the `nlp` pipeline.
<!-- TODO: <Project id="tutorials/ner-relations"> <Project id="tutorials/rel_component">
Run this example use-case by using our project template. It includes all the
</Project> --> code to create the ML model and the pipeline component from scratch.
It also contains two config files to train the model:
one to run on CPU with a Tok2Vec layer, and one for the GPU using a transformer.
The project applies the relation extraction component to identify biomolecular
interactions in a sample dataset, but you can easily swap in your own dataset
for your experiments in any other domain.
</Project>
#### Step 1: Implementing the Model {#component-rel-model} #### Step 1: Implementing the Model {#component-rel-model}
@ -552,41 +560,17 @@ matrix** (~~Floats2d~~) of predictions:
> for details. > for details.
```python ```python
### Register the model architecture ### The model architecture
@registry.architectures.register("rel_model.v1") @spacy.registry.architectures.register("rel_model.v1")
def create_relation_model(...) -> Model[List[Doc], Floats2d]: def create_relation_model(...) -> Model[List[Doc], Floats2d]:
model = ... # 👈 model will go here model = ... # 👈 model will go here
return model return model
``` ```
The first layer in this model will typically be an We adapt a **modular approach** to the definition of this relation model, and
[embedding layer](/usage/embeddings-transformers) such as a define it as chaining two layers together: the first layer that generates an
[`Tok2Vec`](/api/tok2vec) component or a [`Transformer`](/api/transformer). This instance tensor from a given set of documents, and the second layer that
layer is assumed to be of type ~~Model[List[Doc], List[Floats2d]]~~ as it transforms the instance tensor into a final tensor holding the predictions:
transforms each **document into a list of tokens**, with each token being
represented by its embedding in the vector space.
Next, we need a method that **generates pairs of entities** that we want to
classify as being related or not. As these candidate pairs are typically formed
within one document, this function takes a [`Doc`](/api/doc) as input and
outputs a `List` of `Span` tuples. For instance, a very straightforward
implementation would be to just take any two entities from the same document:
```python
### Simple candiate generation
def get_candidates(doc: Doc) -> List[Tuple[Span, Span]]:
candidates = []
for ent1 in doc.ents:
for ent2 in doc.ents:
candidates.append((ent1, ent2))
return candidates
```
But we could also refine this further by **excluding relations** of an entity
with itself, and posing a **maximum distance** (in number of tokens) between two
entities. We register this function in the
[`@misc` registry](/api/top-level#registry) so we can refer to it from the
config, and easily swap it out for any other candidate generation function.
> #### config.cfg (excerpt) > #### config.cfg (excerpt)
> >
@ -594,18 +578,159 @@ config, and easily swap it out for any other candidate generation function.
> [model] > [model]
> @architectures = "rel_model.v1" > @architectures = "rel_model.v1"
> >
> [model.tok2vec] > [model.create_instance_tensor]
> # ... > # ...
> >
> [model.get_candidates] > [model.classification_layer]
> @misc = "rel_cand_generator.v1" > # ...
> max_length = 20
> ``` > ```
```python ```python
### Extended candidate generation {highlight="1,2,7,8"} ### The model architecture {highlight="6"}
@registry.misc.register("rel_cand_generator.v1") @spacy.registry.architectures.register("rel_model.v1")
def create_candidate_indices(max_length: int) -> Callable[[Doc], List[Tuple[Span, Span]]]: def create_relation_model(
create_instance_tensor: Model[List[Doc], Floats2d],
classification_layer: Model[Floats2d, Floats2d],
) -> Model[List[Doc], Floats2d]:
model = chain(create_instance_tensor, classification_layer)
return model
```
The `classification_layer` could be something like a
[Linear](https://thinc.ai/docs/api-layers#linear) layer followed by a
[logistic](https://thinc.ai/docs/api-layers#logistic) activation function:
> #### config.cfg (excerpt)
>
> ```ini
> [model.classification_layer]
> @architectures = "rel_classification_layer.v1"
> nI = null
> nO = null
> ```
```python
### The classification layer
@spacy.registry.architectures.register("rel_classification_layer.v1")
def create_classification_layer(
nO: int = None, nI: int = None
) -> Model[Floats2d, Floats2d]:
return chain(Linear(nO=nO, nI=nI), Logistic())
```
The first layer that **creates the instance tensor** can be defined by
implementing a
[custom forward function](https://thinc.ai/docs/usage-models#weights-layers-forward)
with an appropriate backpropagation callback. We also define an
[initialization method](https://thinc.ai/docs/usage-models#weights-layers-init)
that ensures that the layer is properly set up for training.
We omit some of the implementation details here, and refer to the
[spaCy project](https://github.com/explosion/projects/tree/v3/tutorials/rel_component)
that has the full implementation.
> #### config.cfg (excerpt)
>
> ```ini
> [model.create_instance_tensor]
> @architectures = "rel_instance_tensor.v1"
>
> [model.create_instance_tensor.tok2vec]
> @architectures = "spacy.HashEmbedCNN.v1"
> # ...
>
> [model.create_instance_tensor.pooling]
> @layers = "reduce_mean.v1"
>
> [model.create_instance_tensor.get_instances]
> # ...
> ```
```python
### The layer that creates the instance tensor
@spacy.registry.architectures.register("rel_instance_tensor.v1")
def create_tensors(
tok2vec: Model[List[Doc], List[Floats2d]],
pooling: Model[Ragged, Floats2d],
get_instances: Callable[[Doc], List[Tuple[Span, Span]]],
) -> Model[List[Doc], Floats2d]:
return Model(
"instance_tensors",
instance_forward,
init=instance_init,
layers=[tok2vec, pooling],
refs={"tok2vec": tok2vec, "pooling": pooling},
attrs={"get_instances": get_instances},
)
# The custom forward function
def instance_forward(
model: Model[List[Doc], Floats2d],
docs: List[Doc],
is_train: bool,
) -> Tuple[Floats2d, Callable]:
tok2vec = model.get_ref("tok2vec")
tokvecs, bp_tokvecs = tok2vec(docs, is_train)
get_instances = model.attrs["get_instances"]
all_instances = [get_instances(doc) for doc in docs]
pooling = model.get_ref("pooling")
relations = ...
def backprop(d_relations: Floats2d) -> List[Doc]:
d_tokvecs = ...
return bp_tokvecs(d_tokvecs)
return relations, backprop
# The custom initialization method
def instance_init(
model: Model,
X: List[Doc] = None,
Y: Floats2d = None,
) -> Model:
tok2vec = model.get_ref("tok2vec")
tok2vec.initialize(X)
return model
```
This custom layer uses an [embedding layer](/usage/embeddings-transformers) such
as a [`Tok2Vec`](/api/tok2vec) component or a [`Transformer`](/api/transformer).
This layer is assumed to be of type ~~Model[List[Doc], List[Floats2d]]~~ as it
transforms each **document into a list of tokens**, with each token being
represented by its embedding in the vector space.
The `pooling` layer will be applied to summarize the token vectors into **entity
vectors**, as named entities (represented by ~~Span~~ objects) can consist of
one or multiple tokens. For instance, the pooling layer could resort to
calculating the average of all token vectors in an entity. Thinc provides
several
[built-in pooling operators](https://thinc.ai/docs/api-layers#reduction-ops) for
this purpose.
Finally, we need a `get_instances` method that **generates pairs of entities**
that we want to classify as being related or not. As these candidate pairs are
typically formed within one document, this function takes a [`Doc`](/api/doc) as
input and outputs a `List` of `Span` tuples. For instance, the following
implementation takes any two entities from the same document, as long as they
are within a **maximum distance** (in number of tokens) of eachother:
> #### config.cfg (excerpt)
>
> ```ini
>
> [model.create_instance_tensor.get_instances]
> @misc = "rel_instance_generator.v1"
> max_length = 100
> ```
```python
### Candidate generation
@spacy.registry.misc.register("rel_instance_generator.v1")
def create_instances(max_length: int) -> Callable[[Doc], List[Tuple[Span, Span]]]:
def get_candidates(doc: "Doc") -> List[Tuple[Span, Span]]: def get_candidates(doc: "Doc") -> List[Tuple[Span, Span]]:
candidates = [] candidates = []
for ent1 in doc.ents: for ent1 in doc.ents:
@ -617,45 +742,39 @@ def create_candidate_indices(max_length: int) -> Callable[[Doc], List[Tuple[Span
return get_candidates return get_candidates
``` ```
Finally, we require a method that transforms the candidate entity pairs into a This function in added to the [`@misc` registry](/api/top-level#registry) so we
2D tensor using the specified [`Tok2Vec`](/api/tok2vec) or can refer to it from the config, and easily swap it out for any other candidate
[`Transformer`](/api/transformer). The resulting ~~Floats2~~ object will then be generation function.
processed by a final `output_layer` of the network. Putting all this together,
we can define our relation model in a config file as such:
```ini #### Intermezzo: define how to store the relations data {#component-rel-attribute}
### config.cfg
[model]
@architectures = "rel_model.v1"
# ...
[model.tok2vec] > #### Example output
# ... >
> ```python
> doc = nlp("Amsterdam is the capital of the Netherlands.")
> print("spans", [(e.start, e.text, e.label_) for e in doc.ents])
> for value, rel_dict in doc._.rel.items():
> print(f"{value}: {rel_dict}")
>
> # spans [(0, 'Amsterdam', 'LOC'), (6, 'Netherlands', 'LOC')]
> # (0, 6): {'CAPITAL_OF': 0.89, 'LOCATED_IN': 0.75, 'UNRELATED': 0.002}
> # (6, 0): {'CAPITAL_OF': 0.01, 'LOCATED_IN': 0.13, 'UNRELATED': 0.017}
> ```
[model.get_candidates] For our new relation extraction component, we will use a custom
@misc = "rel_cand_generator.v1" [extension attribute](/usage/processing-pipelines#custom-components-attributes)
max_length = 20 `doc._.rel` in which we store relation data. The attribute refers to a
dictionary, keyed by the **start offsets of each entity** involved in the
[model.create_candidate_tensor] candidate relation. The values in the dictionary refer to another dictionary
@misc = "rel_cand_tensor.v1" where relation labels are mapped to values between 0 and 1. We assume anything
above 0.5 to be a `True` relation. The ~~Example~~ instances that we'll use as
[model.output_layer] training data, will include their gold-standard relation annotations in
@architectures = "rel_output_layer.v1" `example.reference._.rel`.
# ...
```
<!-- TODO: link to project for implementation details -->
<!-- TODO: maybe embed files from project that show the architectures? -->
When creating this model, we store the custom functions as
[attributes](https://thinc.ai/docs/api-model#properties) and the sublayers as
references, so we can access them easily:
```python ```python
tok2vec_layer = model.get_ref("tok2vec") ### Registering the extension attribute
output_layer = model.get_ref("output_layer") from spacy.tokens import Doc
create_candidate_tensor = model.attrs["create_candidate_tensor"] Doc.set_extension("rel", default={})
get_candidates = model.attrs["get_candidates"]
``` ```
#### Step 2: Implementing the pipeline component {#component-rel-pipe} #### Step 2: Implementing the pipeline component {#component-rel-pipe}
@ -698,19 +817,44 @@ class RelationExtractor(TrainablePipe):
... ...
``` ```
Before the model can be used, it needs to be Typically, the **constructor** defines the vocab, the Machine Learning model,
[initialized](/usage/training#initialization). This function receives a callback and the name of this component. Additionally, this component, just like the
to access the full **training data set**, or a representative sample. This data `textcat` and the `tagger`, stores an **internal list of labels**. The ML model
set can be used to deduce all **relevant labels**. Alternatively, a list of will predict scores for each label. We add convenience methods to easily
labels can be provided to `initialize`, or you can call retrieve and add to them.
`RelationExtractor.add_label` directly. The number of labels defines the output
dimensionality of the network, and will be used to do ```python
### The constructor (continued)
def __init__(self, vocab, model, name="rel"):
"""Create a component instance."""
# ...
self.cfg = {"labels": []}
@property
def labels(self) -> Tuple[str]:
"""Returns the labels currently added to the component."""
return tuple(self.cfg["labels"])
def add_label(self, label: str):
"""Add a new label to the pipe."""
self.cfg["labels"] = list(self.labels) + [label]
```
After creation, the component needs to be
[initialized](/usage/training#initialization). This method can define the
relevant labels in two ways: explicitely by setting the `labels` argument in the
[`initialize` block](/api/data-formats#config-initialize) of the config, or
implicately by deducing them from the `get_examples` callback that generates the
full **training data set**, or a representative sample.
The final number of labels defines the output dimensionality of the network, and
will be used to do
[shape inference](https://thinc.ai/docs/usage-models#validation) throughout the [shape inference](https://thinc.ai/docs/usage-models#validation) throughout the
layers of the neural network. This is triggered by calling layers of the neural network. This is triggered by calling
[`Model.initialize`](https://thinc.ai/api/model#initialize). [`Model.initialize`](https://thinc.ai/api/model#initialize).
```python ```python
### The initialize method {highlight="12,18,22"} ### The initialize method {highlight="12,15,18,22"}
from itertools import islice from itertools import islice
def initialize( def initialize(
@ -741,7 +885,7 @@ Typically, this happens when the pipeline is set up before training in
[`spacy train`](/api/cli#training). After initialization, the pipeline component [`spacy train`](/api/cli#training). After initialization, the pipeline component
and its internal model can be trained and used to make predictions. and its internal model can be trained and used to make predictions.
During training, the function [`update`](/api/pipe#update) is invoked which During training, the method [`update`](/api/pipe#update) is invoked which
delegates to delegates to
[`Model.begin_update`](https://thinc.ai/docs/api-model#begin_update) and a [`Model.begin_update`](https://thinc.ai/docs/api-model#begin_update) and a
[`get_loss`](/api/pipe#get_loss) function that **calculates the loss** for a [`get_loss`](/api/pipe#get_loss) function that **calculates the loss** for a
@ -761,18 +905,18 @@ def update(
sgd: Optional[Optimizer] = None, sgd: Optional[Optimizer] = None,
losses: Optional[Dict[str, float]] = None, losses: Optional[Dict[str, float]] = None,
) -> Dict[str, float]: ) -> Dict[str, float]:
... # ...
docs = [ex.predicted for ex in examples] docs = [eg.predicted for eg in examples]
predictions, backprop = self.model.begin_update(docs) predictions, backprop = self.model.begin_update(docs)
loss, gradient = self.get_loss(examples, predictions) loss, gradient = self.get_loss(examples, predictions)
backprop(gradient) backprop(gradient)
losses[self.name] += loss losses[self.name] += loss
... # ...
return losses return losses
``` ```
When the internal model is trained, the component can be used to make novel After training the model, the component can be used to make novel
**predictions**. The [`predict`](/api/pipe#predict) function needs to be **predictions**. The [`predict`](/api/pipe#predict) method needs to be
implemented for each subclass of `TrainablePipe`. In our case, we can simply implemented for each subclass of `TrainablePipe`. In our case, we can simply
delegate to the internal model's delegate to the internal model's
[predict](https://thinc.ai/docs/api-model#predict) function that takes a batch [predict](https://thinc.ai/docs/api-model#predict) function that takes a batch
@ -788,42 +932,21 @@ def predict(self, docs: Iterable[Doc]) -> Floats2d:
The final method that needs to be implemented, is The final method that needs to be implemented, is
[`set_annotations`](/api/pipe#set_annotations). This function takes the [`set_annotations`](/api/pipe#set_annotations). This function takes the
predictions, and modifies the given `Doc` object in place to store them. For our predictions, and modifies the given `Doc` object in place to store them. For our
relation extraction component, we store the data as a dictionary in a custom relation extraction component, we store the data in the
[extension attribute](/usage/processing-pipelines#custom-components-attributes) [custom attribute](#component-rel-attribute)`doc._.rel`.
`doc._.rel`. As keys, we represent the candidate pair by the **start offsets of
each entity**, as this defines an entity pair uniquely within one document.
To interpret the scores predicted by the relation extraction model correctly, we To interpret the scores predicted by the relation extraction model correctly, we
need to refer to the model's `get_candidates` function that defined which pairs need to refer to the model's `get_instances` function that defined which pairs
of entities were relevant candidates, so that the predictions can be linked to of entities were relevant candidates, so that the predictions can be linked to
those exact entities: those exact entities:
> #### Example output
>
> ```python
> doc = nlp("Amsterdam is the capital of the Netherlands.")
> print("spans", [(e.start, e.text, e.label_) for e in doc.ents])
> for value, rel_dict in doc._.rel.items():
> print(f"{value}: {rel_dict}")
>
> # spans [(0, 'Amsterdam', 'LOC'), (6, 'Netherlands', 'LOC')]
> # (0, 6): {'CAPITAL_OF': 0.89, 'LOCATED_IN': 0.75, 'UNRELATED': 0.002}
> # (6, 0): {'CAPITAL_OF': 0.01, 'LOCATED_IN': 0.13, 'UNRELATED': 0.017}
> ```
```python
### Registering the extension attribute
from spacy.tokens import Doc
Doc.set_extension("rel", default={})
```
```python ```python
### The set_annotations method {highlight="5-6,10"} ### The set_annotations method {highlight="5-6,10"}
def set_annotations(self, docs: Iterable[Doc], predictions: Floats2d): def set_annotations(self, docs: Iterable[Doc], predictions: Floats2d):
c = 0 c = 0
get_candidates = self.model.attrs["get_candidates"] get_instances = self.model.attrs["get_instances"]
for doc in docs: for doc in docs:
for (e1, e2) in get_candidates(doc): for (e1, e2) in get_instances(doc):
offset = (e1.start, e2.start) offset = (e1.start, e2.start)
if offset not in doc._.rel: if offset not in doc._.rel:
doc._.rel[offset] = {} doc._.rel[offset] = {}
@ -837,15 +960,15 @@ Under the hood, when the pipe is applied to a document, it delegates to the
```python ```python
### The __call__ method ### The __call__ method
def __call__(self, Doc doc): def __call__(self, doc: Doc):
predictions = self.predict([doc]) predictions = self.predict([doc])
self.set_annotations([doc], predictions) self.set_annotations([doc], predictions)
return doc return doc
``` ```
There is one more optional method to implement: [`score`](/api/pipe#score) There is one more optional method to implement: [`score`](/api/pipe#score)
calculates the performance of your component on a set of examples, and calculates the performance of your component on a set of examples, and returns
returns the results as a dictionary: the results as a dictionary:
```python ```python
### The score method ### The score method
@ -861,8 +984,8 @@ def score(self, examples: Iterable[Example]) -> Dict[str, Any]:
} }
``` ```
This is particularly useful to see the scores on the development corpus This is particularly useful for calculating relevant scores on the development
when training the component with [`spacy train`](/api/cli#training). corpus when training the component with [`spacy train`](/api/cli#training).
Once our `TrainablePipe` subclass is fully implemented, we can Once our `TrainablePipe` subclass is fully implemented, we can
[register](/usage/processing-pipelines#custom-components-factories) the [register](/usage/processing-pipelines#custom-components-factories) the
@ -879,14 +1002,8 @@ assigns it a name and lets you create the component with
> >
> [components.relation_extractor.model] > [components.relation_extractor.model]
> @architectures = "rel_model.v1" > @architectures = "rel_model.v1"
>
> [components.relation_extractor.model.tok2vec]
> # ... > # ...
> >
> [components.relation_extractor.model.get_candidates]
> @misc = "rel_cand_generator.v1"
> max_length = 20
>
> [training.score_weights] > [training.score_weights]
> rel_micro_p = 0.0 > rel_micro_p = 0.0
> rel_micro_r = 0.0 > rel_micro_r = 0.0
@ -902,8 +1019,8 @@ def make_relation_extractor(nlp, name, model):
return RelationExtractor(nlp.vocab, model, name) return RelationExtractor(nlp.vocab, model, name)
``` ```
You can extend the decorator to include information such as the type of You can extend the decorator to include information such as the type of
annotations that are required for this component to run, the type of annotations annotations that are required for this component to run, the type of annotations
it produces, and the scores that can be calculated: it produces, and the scores that can be calculated:
```python ```python
@ -924,6 +1041,12 @@ def make_relation_extractor(nlp, name, model):
return RelationExtractor(nlp.vocab, model, name) return RelationExtractor(nlp.vocab, model, name)
``` ```
<!-- TODO: <Project id="tutorials/ner-relations"> <Project id="tutorials/rel_component">
Run this example use-case by using our project template. It includes all the
</Project> --> code to create the ML model and the pipeline component from scratch.
It contains two config files to train the model:
one to run on CPU with a Tok2Vec layer, and one for the GPU using a transformer.
The project applies the relation extraction component to identify biomolecular
interactions, but you can easily swap in your own dataset for your experiments
in any other domain.
</Project>