Fix use_gold_ents behaviour for EntityLinker (#13400)

* fix type annotation in docs

* only restore entities after loss calculation

* restore entities of sample in initialization

* rename overfitting function

* fix EL scorer

* Relax test

* fix formatting

* Update spacy/pipeline/entity_linker.py

Co-authored-by: Raphael Mitsch <[email protected]>

* rename to _ensure_ents

* further rename

* allow for scorer to be None

---------

Co-authored-by: Raphael Mitsch <[email protected]>

Author: svlandeg

spacy/pipeline/entity_linker.py

@@ -11,7 +11,6 @@
 from ..errors import Errors
 from ..kb import Candidate, KnowledgeBase
 from ..language import Language
-from ..ml import empty_kb
 from ..scorer import Scorer
 from ..tokens import Doc, Span
 from ..training import Example, validate_examples, validate_get_examples
@@ -105,7 +104,7 @@ def make_entity_linker(
         ): Function that produces a list of candidates, given a certain knowledge base and several textual mentions.
     generate_empty_kb (Callable[[Vocab, int], KnowledgeBase]): Callable returning empty KnowledgeBase.
     scorer (Optional[Callable]): The scoring method.
-    use_gold_ents (bool): Whether to copy entities from gold docs or not. If false, another
+    use_gold_ents (bool): Whether to copy entities from gold docs during training or not. If false, another
         component must provide entity annotations.
     candidates_batch_size (int): Size of batches for entity candidate generation.
     threshold (Optional[float]): Confidence threshold for entity predictions. If confidence is below the threshold,
@@ -235,14 +234,44 @@ def __init__(
         self.cfg: Dict[str, Any] = {"overwrite": overwrite}
         self.distance = CosineDistance(normalize=False)
         self.kb = generate_empty_kb(self.vocab, entity_vector_length)
-        self.scorer = scorer
         self.use_gold_ents = use_gold_ents
         self.candidates_batch_size = candidates_batch_size
         self.threshold = threshold
 
         if candidates_batch_size < 1:
             raise ValueError(Errors.E1044)
 
+        def _score_with_ents_set(examples: Iterable[Example], **kwargs):
+            # Because of how spaCy works, we can't just score immediately, because Language.evaluate
+            # calls pipe() on the predicted docs, which won't have entities if there is no NER in the pipeline.
+            if not scorer:
+                return scorer
+            if not self.use_gold_ents:
+                return scorer(examples, **kwargs)
+            else:
+                examples = self._ensure_ents(examples)
+                docs = self.pipe(
+                    (eg.predicted for eg in examples),
+                )
+                for eg, doc in zip(examples, docs):
+                    eg.predicted = doc
+                return scorer(examples, **kwargs)
+
+        self.scorer = _score_with_ents_set
+
+    def _ensure_ents(self, examples: Iterable[Example]) -> Iterable[Example]:
+        """If use_gold_ents is true, set the gold entities to (a copy of) eg.predicted."""
+        if not self.use_gold_ents:
+            return examples
+
+        new_examples = []
+        for eg in examples:
+            ents, _ = eg.get_aligned_ents_and_ner()
+            new_eg = eg.copy()
+            new_eg.predicted.ents = ents
+            new_examples.append(new_eg)
+        return new_examples
+
     def set_kb(self, kb_loader: Callable[[Vocab], KnowledgeBase]):
         """Define the KB of this pipe by providing a function that will
         create it using this object's vocab."""
@@ -284,11 +313,9 @@ def initialize(
         nO = self.kb.entity_vector_length
         doc_sample = []
         vector_sample = []
-        for eg in islice(get_examples(), 10):
+        examples = self._ensure_ents(islice(get_examples(), 10))
+        for eg in examples:
             doc = eg.x
-            if self.use_gold_ents:
-                ents, _ = eg.get_aligned_ents_and_ner()
-                doc.ents = ents
             doc_sample.append(doc)
             vector_sample.append(self.model.ops.alloc1f(nO))
         assert len(doc_sample) > 0, Errors.E923.format(name=self.name)
@@ -354,43 +381,31 @@ def update(
         losses.setdefault(self.name, 0.0)
         if not examples:
             return losses
+        examples = self._ensure_ents(examples)
         validate_examples(examples, "EntityLinker.update")
 
-        set_dropout_rate(self.model, drop)
-        docs = [eg.predicted for eg in examples]
-        # save to restore later
-        old_ents = [doc.ents for doc in docs]
-
-        for doc, ex in zip(docs, examples):
-            if self.use_gold_ents:
-                ents, _ = ex.get_aligned_ents_and_ner()
-                doc.ents = ents
-            else:
-                # only keep matching ents
-                doc.ents = ex.get_matching_ents()
-
         # make sure we have something to learn from, if not, short-circuit
         if not self.batch_has_learnable_example(examples):
             return losses
 
+        set_dropout_rate(self.model, drop)
+        docs = [eg.predicted for eg in examples]
         sentence_encodings, bp_context = self.model.begin_update(docs)
 
-        # now restore the ents
-        for doc, old in zip(docs, old_ents):
-            doc.ents = old
-
         loss, d_scores = self.get_loss(
             sentence_encodings=sentence_encodings, examples=examples
         )
         bp_context(d_scores)
         if sgd is not None:
             self.finish_update(sgd)
         losses[self.name] += loss
+
         return losses
 
     def get_loss(self, examples: Iterable[Example], sentence_encodings: Floats2d):
         validate_examples(examples, "EntityLinker.get_loss")
         entity_encodings = []
+        # We assume that get_loss is called with gold ents set in the examples if need be
         eidx = 0  # indices in gold entities to keep
         keep_ents = []  # indices in sentence_encodings to keep
 

spacy/tests/pipeline/test_entity_linker.py

@@ -717,7 +717,7 @@ def test_preserving_links_ents_2(nlp):
 # fmt: on
 
 
-def test_overfitting_IO():
+def test_overfitting_IO_gold_entities():
     # Simple test to try and quickly overfit the NEL component - ensuring the ML models work correctly
     nlp = English()
     vector_length = 3
@@ -744,7 +744,9 @@ def create_kb(vocab):
         return mykb
 
     # Create the Entity Linker component and add it to the pipeline
-    entity_linker = nlp.add_pipe("entity_linker", last=True)
+    entity_linker = nlp.add_pipe(
+        "entity_linker", last=True, config={"use_gold_ents": True}
+    )
     assert isinstance(entity_linker, EntityLinker)
     entity_linker.set_kb(create_kb)
     assert "Q2146908" in entity_linker.vocab.strings
@@ -807,6 +809,107 @@ def create_kb(vocab):
     assert_equal(batch_deps_1, batch_deps_2)
     assert_equal(batch_deps_1, no_batch_deps)
 
+    eval = nlp.evaluate(train_examples)
+    assert "nel_macro_p" in eval
+    assert "nel_macro_r" in eval
+    assert "nel_macro_f" in eval
+    assert "nel_micro_p" in eval
+    assert "nel_micro_r" in eval
+    assert "nel_micro_f" in eval
+    assert "nel_f_per_type" in eval
+    assert "PERSON" in eval["nel_f_per_type"]
+
+    assert eval["nel_macro_f"] > 0
+    assert eval["nel_micro_f"] > 0
+
+
+def test_overfitting_IO_with_ner():
+    # Simple test to try and overfit the NER and NEL component in combination - ensuring the ML models work correctly
+    nlp = English()
+    vector_length = 3
+    assert "Q2146908" not in nlp.vocab.strings
+
+    # Convert the texts to docs to make sure we have doc.ents set for the training examples
+    train_examples = []
+    for text, annotation in TRAIN_DATA:
+        doc = nlp(text)
+        train_examples.append(Example.from_dict(doc, annotation))
+
+    def create_kb(vocab):
+        # create artificial KB - assign same prior weight to the two russ cochran's
+        # Q2146908 (Russ Cochran): American golfer
+        # Q7381115 (Russ Cochran): publisher
+        mykb = InMemoryLookupKB(vocab, entity_vector_length=vector_length)
+        mykb.add_entity(entity="Q2146908", freq=12, entity_vector=[6, -4, 3])
+        mykb.add_entity(entity="Q7381115", freq=12, entity_vector=[9, 1, -7])
+        mykb.add_alias(
+            alias="Russ Cochran",
+            entities=["Q2146908", "Q7381115"],
+            probabilities=[0.5, 0.5],
+        )
+        return mykb
+
+    # Create the NER and EL components and add them to the pipeline
+    ner = nlp.add_pipe("ner", first=True)
+    entity_linker = nlp.add_pipe(
+        "entity_linker", last=True, config={"use_gold_ents": False}
+    )
+    entity_linker.set_kb(create_kb)
+
+    train_examples = []
+    for text, annotations in TRAIN_DATA:
+        train_examples.append(Example.from_dict(nlp.make_doc(text), annotations))
+        for ent in annotations.get("entities"):
+            ner.add_label(ent[2])
+    optimizer = nlp.initialize()
+
+    # train the NER and NEL pipes
+    for i in range(50):
+        losses = {}
+        nlp.update(train_examples, sgd=optimizer, losses=losses)
+    assert losses["ner"] < 0.001
+    assert losses["entity_linker"] < 0.001
+
+    # adding additional components that are required for the entity_linker
+    nlp.add_pipe("sentencizer", first=True)
+
+    # test the trained model
+    test_text = "Russ Cochran captured his first major title with his son as caddie."
+    doc = nlp(test_text)
+    ents = doc.ents
+    assert len(ents) == 1
+    assert ents[0].text == "Russ Cochran"
+    assert ents[0].label_ == "PERSON"
+    assert ents[0].kb_id_ != "NIL"
+
+    # TODO: below assert is still flaky - EL doesn't properly overfit quite yet
+    # assert ents[0].kb_id_ == "Q2146908"
+
+    # Also test the results are still the same after IO
+    with make_tempdir() as tmp_dir:
+        nlp.to_disk(tmp_dir)
+        nlp2 = util.load_model_from_path(tmp_dir)
+        assert nlp2.pipe_names == nlp.pipe_names
+        doc2 = nlp2(test_text)
+        ents2 = doc2.ents
+        assert len(ents2) == 1
+        assert ents2[0].text == "Russ Cochran"
+        assert ents2[0].label_ == "PERSON"
+        assert ents2[0].kb_id_ != "NIL"
+
+    eval = nlp.evaluate(train_examples)
+    assert "nel_macro_f" in eval
+    assert "nel_micro_f" in eval
+    assert "ents_f" in eval
+    assert "nel_f_per_type" in eval
+    assert "ents_per_type" in eval
+    assert "PERSON" in eval["nel_f_per_type"]
+    assert "PERSON" in eval["ents_per_type"]
+
+    assert eval["nel_macro_f"] > 0
+    assert eval["nel_micro_f"] > 0
+    assert eval["ents_f"] > 0
+
 
 def test_kb_serialization():
     # Test that the KB can be used in a pipeline with a different vocab

website/docs/api/entitylinker.mdx

@@ -61,7 +61,7 @@ architectures and their arguments and hyperparameters.
 | `incl_context`                                      | Whether or not to include the local context in the model. Defaults to `True`. ~~bool~~                                                                                                                                                                                                                           |
 | `model`                                             | The [`Model`](https://thinc.ai/docs/api-model) powering the pipeline component. Defaults to [EntityLinker](/api/architectures#EntityLinker). ~~Model~~                                                                                                                                                           |
 | `entity_vector_length`                              | Size of encoding vectors in the KB. Defaults to `64`. ~~int~~                                                                                                                                                                                                                                                    |
-| `use_gold_ents`                                     | Whether to copy entities from the gold docs or not. Defaults to `True`. If `False`, entities must be set in the training data or by an annotating component in the pipeline. ~~int~~                                                                                                                             |
+| `use_gold_ents`                                     | Whether to copy entities from the gold docs or not. Defaults to `True`. If `False`, entities must be set in the training data or by an annotating component in the pipeline. ~~bool~~                                                                                                                            |
 | `get_candidates`                                    | Function that generates plausible candidates for a given `Span` object. Defaults to [CandidateGenerator](/api/architectures#CandidateGenerator), a function looking up exact, case-dependent aliases in the KB. ~~Callable[[KnowledgeBase, Span], Iterable[Candidate]]~~                                         |
 | `get_candidates_batch` <Tag variant="new">3.5</Tag> | Function that generates plausible candidates for a given batch of `Span` objects. Defaults to [CandidateBatchGenerator](/api/architectures#CandidateBatchGenerator), a function looking up exact, case-dependent aliases in the KB. ~~Callable[[KnowledgeBase, Iterable[Span]], Iterable[Iterable[Candidate]]]~~ |
 | `generate_empty_kb` <Tag variant="new">3.5.1</Tag>  | Function that generates an empty `KnowledgeBase` object. Defaults to [`spacy.EmptyKB.v2`](/api/architectures#EmptyKB), which generates an empty [`InMemoryLookupKB`](/api/inmemorylookupkb). ~~Callable[[Vocab, int], KnowledgeBase]~~                                                                           |