Store activations in Doc when store_activations is enabled

This change adds the new `activations` attribute to `Doc`. This
attribute can be used by trainable pipes to store their activations,
probabilities, and guesses for downstream users.

As an example, this change modifies the `tagger` and `senter` pipes to
add an `store_activations` option. When this option is enabled, the
probabilities and guesses are stored in `set_annotations`.

spacy/pipeline/morphologizer.pyx

@@ -217,7 +217,7 @@ class Morphologizer(Tagger):
         assert len(label_sample) > 0, Errors.E923.format(name=self.name)
         self.model.initialize(X=doc_sample, Y=label_sample)
 
-    def set_annotations(self, docs, batch_tag_ids):
+    def set_annotations(self, docs, scores_guesses):
         """Modify a batch of documents, using pre-computed scores.
 
         docs (Iterable[Doc]): The documents to modify.
@@ -225,6 +225,7 @@ class Morphologizer(Tagger):
 
         DOCS: https://spacy.io/api/morphologizer#set_annotations
         """
+        _, batch_tag_ids = scores_guesses
         if isinstance(docs, Doc):
             docs = [docs]
         cdef Doc doc

spacy/pipeline/senter.pyx

@@ -38,11 +38,16 @@ DEFAULT_SENTER_MODEL = Config().from_str(default_model_config)["model"]
 @Language.factory(
     "senter",
     assigns=["token.is_sent_start"],
-    default_config={"model": DEFAULT_SENTER_MODEL, "overwrite": False, "scorer": {"@scorers": "spacy.senter_scorer.v1"}},
+    default_config={
+        "model": DEFAULT_SENTER_MODEL,
+        "overwrite": False,
+        "scorer": {"@scorers": "spacy.senter_scorer.v1"},
+        "store_activations": False
+    },
     default_score_weights={"sents_f": 1.0, "sents_p": 0.0, "sents_r": 0.0},
 )
-def make_senter(nlp: Language, name: str, model: Model, overwrite: bool, scorer: Optional[Callable]):
+def make_senter(nlp: Language, name: str, model: Model, overwrite: bool, scorer: Optional[Callable], store_activations: bool):
-    return SentenceRecognizer(nlp.vocab, model, name, overwrite=overwrite, scorer=scorer)
+    return SentenceRecognizer(nlp.vocab, model, name, overwrite=overwrite, scorer=scorer, store_activations=store_activations)
 
 
 def senter_score(examples, **kwargs):
@@ -72,6 +77,7 @@ class SentenceRecognizer(Tagger):
         *,
         overwrite=BACKWARD_OVERWRITE,
         scorer=senter_score,
+        store_activations=False,
     ):
         """Initialize a sentence recognizer.
 
@@ -90,6 +96,7 @@ class SentenceRecognizer(Tagger):
         self._rehearsal_model = None
         self.cfg = {"overwrite": overwrite}
         self.scorer = scorer
+        self.store_activations = store_activations
 
     @property
     def labels(self):
@@ -107,7 +114,7 @@ class SentenceRecognizer(Tagger):
     def label_data(self):
         return None
 
-    def set_annotations(self, docs, batch_tag_ids):
+    def set_annotations(self, docs, scores_guesses):
         """Modify a batch of documents, using pre-computed scores.
 
         docs (Iterable[Doc]): The documents to modify.
@@ -115,11 +122,17 @@ class SentenceRecognizer(Tagger):
 
         DOCS: https://spacy.io/api/sentencerecognizer#set_annotations
         """
+        _, batch_tag_ids = scores_guesses
         if isinstance(docs, Doc):
             docs = [docs]
         cdef Doc doc
         cdef bint overwrite = self.cfg["overwrite"]
         for i, doc in enumerate(docs):
+            if self.store_activations:
+                doc.activations[self.name] = {
+                    "probs": scores_guesses[0][i],
+                    "guesses": scores_guesses[1][i],
+                }
             doc_tag_ids = batch_tag_ids[i]
             if hasattr(doc_tag_ids, "get"):
                 doc_tag_ids = doc_tag_ids.get()

spacy/pipeline/tagger.pyx

@@ -45,7 +45,13 @@ DEFAULT_TAGGER_MODEL = Config().from_str(default_model_config)["model"]
 @Language.factory(
     "tagger",
     assigns=["token.tag"],
-    default_config={"model": DEFAULT_TAGGER_MODEL, "overwrite": False, "scorer": {"@scorers": "spacy.tagger_scorer.v1"}, "neg_prefix": "!"},
+    default_config={
+        "model": DEFAULT_TAGGER_MODEL,
+        "overwrite": False,
+        "scorer": {"@scorers": "spacy.tagger_scorer.v1"},
+        "neg_prefix": "!",
+        "store_activations": False
+    },
     default_score_weights={"tag_acc": 1.0},
 )
 def make_tagger(
@@ -55,6 +61,7 @@ def make_tagger(
     overwrite: bool,
     scorer: Optional[Callable],
     neg_prefix: str,
+    store_activations: bool,
 ):
     """Construct a part-of-speech tagger component.
 
@@ -63,7 +70,7 @@ def make_tagger(
         in size, and be normalized as probabilities (all scores between 0 and 1,
         with the rows summing to 1).
     """
-    return Tagger(nlp.vocab, model, name, overwrite=overwrite, scorer=scorer, neg_prefix=neg_prefix)
+    return Tagger(nlp.vocab, model, name, overwrite=overwrite, scorer=scorer, neg_prefix=neg_prefix, store_activations=store_activations)
 
 
 def tagger_score(examples, **kwargs):
@@ -89,6 +96,7 @@ class Tagger(TrainablePipe):
         overwrite=BACKWARD_OVERWRITE,
         scorer=tagger_score,
         neg_prefix="!",
+        store_activations=False,
     ):
         """Initialize a part-of-speech tagger.
 
@@ -108,6 +116,7 @@ class Tagger(TrainablePipe):
         cfg = {"labels": [], "overwrite": overwrite, "neg_prefix": neg_prefix}
         self.cfg = dict(sorted(cfg.items()))
         self.scorer = scorer
+        self.store_activations = store_activations
 
     @property
     def labels(self):
@@ -139,12 +148,12 @@ class Tagger(TrainablePipe):
             n_labels = len(self.labels)
             guesses = [self.model.ops.alloc((0, n_labels)) for doc in docs]
             assert len(guesses) == len(docs)
-            return guesses
+            return guesses, guesses
         scores = self.model.predict(docs)
         assert len(scores) == len(docs), (len(scores), len(docs))
         guesses = self._scores2guesses(scores)
         assert len(guesses) == len(docs)
-        return guesses
+        return scores, guesses
 
     def _scores2guesses(self, scores):
         guesses = []
@@ -155,7 +164,7 @@ class Tagger(TrainablePipe):
             guesses.append(doc_guesses)
         return guesses
 
-    def set_annotations(self, docs, batch_tag_ids):
+    def set_annotations(self, docs, scores_guesses):
         """Modify a batch of documents, using pre-computed scores.
 
         docs (Iterable[Doc]): The documents to modify.
@@ -163,6 +172,7 @@ class Tagger(TrainablePipe):
 
         DOCS: https://spacy.io/api/tagger#set_annotations
         """
+        _, batch_tag_ids = scores_guesses
         if isinstance(docs, Doc):
             docs = [docs]
         cdef Doc doc
@@ -170,6 +180,11 @@ class Tagger(TrainablePipe):
         cdef bint overwrite = self.cfg["overwrite"]
         labels = self.labels
         for i, doc in enumerate(docs):
+            if self.store_activations:
+                doc.activations[self.name] = {
+                    "probs": scores_guesses[0][i],
+                    "guesses": scores_guesses[1][i],
+                }
             doc_tag_ids = batch_tag_ids[i]
             if hasattr(doc_tag_ids, "get"):
                 doc_tag_ids = doc_tag_ids.get()

spacy/pipeline/trainable_pipe.pxd

@@ -6,3 +6,4 @@ cdef class TrainablePipe(Pipe):
     cdef public object model
     cdef public object cfg
     cdef public object scorer
+    cdef public bint store_activations

spacy/tests/pipeline/test_senter.py

@@ -1,3 +1,4 @@
+from typing import cast
 import pytest
 from numpy.testing import assert_equal
 from spacy.attrs import SENT_START
@@ -6,6 +7,7 @@ from spacy import util
 from spacy.training import Example
 from spacy.lang.en import English
 from spacy.language import Language
+from spacy.pipeline import TrainablePipe
 from spacy.tests.util import make_tempdir
 
 
@@ -101,3 +103,23 @@ def test_overfitting_IO():
     # test internal pipe labels vs. Language.pipe_labels with hidden labels
     assert nlp.get_pipe("senter").labels == ("I", "S")
     assert "senter" not in nlp.pipe_labels
+
+
+def test_store_activations():
+    # Simple test to try and quickly overfit the senter - ensuring the ML models work correctly
+    nlp = English()
+    senter = cast(TrainablePipe, nlp.add_pipe("senter"))
+
+    train_examples = []
+    for t in TRAIN_DATA:
+        train_examples.append(Example.from_dict(nlp.make_doc(t[0]), t[1]))
+
+    nlp.initialize(get_examples=lambda: train_examples)
+
+    senter.store_activations = True
+
+    doc = nlp("This is a test.")
+    assert "senter" in doc.activations
+    assert set(doc.activations["senter"].keys()) == {"guesses", "probs"}
+    assert doc.activations["senter"]["probs"].shape == (5, 2)
+    assert doc.activations["senter"]["guesses"].shape == (5,)

spacy/tests/pipeline/test_tagger.py

@@ -1,3 +1,4 @@
+from typing import cast
 import pytest
 from numpy.testing import assert_equal
 from spacy.attrs import TAG
@@ -6,6 +7,7 @@ from spacy import util
 from spacy.training import Example
 from spacy.lang.en import English
 from spacy.language import Language
+from spacy.pipeline import TrainablePipe
 from thinc.api import compounding
 
 from ..util import make_tempdir
@@ -211,6 +213,25 @@ def test_overfitting_IO():
     assert doc3[0].tag_ != "N"
 
 
+def test_store_activations():
+    # Simple test to try and quickly overfit the tagger - ensuring the ML models work correctly
+    nlp = English()
+    tagger = cast(TrainablePipe, nlp.add_pipe("tagger"))
+    train_examples = []
+    for t in TRAIN_DATA:
+        train_examples.append(Example.from_dict(nlp.make_doc(t[0]), t[1]))
+    nlp.initialize(get_examples=lambda: train_examples)
+
+    tagger.store_activations = True
+
+    doc = nlp("This is a test.")
+
+    assert "tagger" in doc.activations
+    assert set(doc.activations["tagger"].keys()) == {"guesses", "probs"}
+    assert doc.activations["tagger"]["probs"].shape == (5, len(TAGS))
+    assert doc.activations["tagger"]["guesses"].shape == (5,)
+
+
 def test_tagger_requires_labels():
     nlp = English()
     nlp.add_pipe("tagger")

spacy/tokens/doc.pxd

@@ -50,6 +50,8 @@ cdef class Doc:
 
     cdef public float sentiment
 
+    cdef public dict activations
+
     cdef public dict user_hooks
     cdef public dict user_token_hooks
     cdef public dict user_span_hooks

spacy/tokens/doc.pyx

@@ -245,6 +245,7 @@ cdef class Doc:
         self.length = 0
         self.sentiment = 0.0
         self.cats = {}
+        self.activations = {}
         self.user_hooks = {}
         self.user_token_hooks = {}
         self.user_span_hooks = {}