Update thinc dependency to 9.0.0.dev4

pyproject.toml

@@ -5,7 +5,7 @@ requires = [
     "cymem>=2.0.2,<2.1.0",
     "preshed>=3.0.2,<3.1.0",
     "murmurhash>=0.28.0,<1.1.0",
-    "thinc>=9.0.0.dev2,<9.1.0",
+    "thinc>=9.0.0.dev4,<9.1.0",
     "numpy>=1.15.0",
 ]
 build-backend = "setuptools.build_meta"

requirements.txt

@@ -3,7 +3,7 @@ spacy-legacy>=4.0.0.dev0,<4.1.0
 spacy-loggers>=1.0.0,<2.0.0
 cymem>=2.0.2,<2.1.0
 preshed>=3.0.2,<3.1.0
-thinc>=9.0.0.dev2,<9.1.0
+thinc>=9.0.0.dev4,<9.1.0
 ml_datasets>=0.2.0,<0.3.0
 murmurhash>=0.28.0,<1.1.0
 wasabi>=0.9.1,<1.2.0

setup.cfg

@@ -30,14 +30,22 @@ project_urls =
 zip_safe = false
 include_package_data = true
 python_requires = >=3.8
+setup_requires =
+    cython>=0.25,<3.0
+    numpy>=1.15.0
+    # We also need our Cython packages here to compile against
+    cymem>=2.0.2,<2.1.0
+    preshed>=3.0.2,<3.1.0
+    murmurhash>=0.28.0,<1.1.0
+    thinc>=9.0.0.dev4,<9.1.0
 install_requires =
     # Our libraries
     spacy-legacy>=4.0.0.dev0,<4.1.0
     spacy-loggers>=1.0.0,<2.0.0
     murmurhash>=0.28.0,<1.1.0
     cymem>=2.0.2,<2.1.0
     preshed>=3.0.2,<3.1.0
-    thinc>=9.0.0.dev2,<9.1.0
+    thinc>=9.0.0.dev4,<9.1.0
     wasabi>=0.9.1,<1.2.0
     srsly>=2.4.3,<3.0.0
     catalogue>=2.0.6,<2.1.0

spacy/pipeline/edit_tree_lemmatizer.py

@@ -5,7 +5,6 @@
 import numpy as np
 import srsly
 from thinc.api import Config, Model, NumpyOps, SequenceCategoricalCrossentropy
-from thinc.legacy import LegacySequenceCategoricalCrossentropy
 from thinc.types import ArrayXd, Floats2d, Ints1d
 
 from .. import util
@@ -128,9 +127,7 @@ def get_loss(
         self, examples: Iterable[Example], scores: List[Floats2d]
     ) -> Tuple[float, List[Floats2d]]:
         validate_examples(examples, "EditTreeLemmatizer.get_loss")
-        loss_func = LegacySequenceCategoricalCrossentropy(
-            normalize=False, missing_value=-1
-        )
+        loss_func = SequenceCategoricalCrossentropy(normalize=False, missing_value=-1)
 
         truths = []
         for eg in examples:
@@ -166,7 +163,7 @@ def get_teacher_student_loss(
 
         DOCS: https://spacy.io/api/edittreelemmatizer#get_teacher_student_loss
         """
-        loss_func = LegacySequenceCategoricalCrossentropy(normalize=False)
+        loss_func = SequenceCategoricalCrossentropy(normalize=False)
         d_scores, loss = loss_func(student_scores, teacher_scores)
         if self.model.ops.xp.isnan(loss):
             raise ValueError(Errors.E910.format(name=self.name))

spacy/pipeline/morphologizer.pyx

@@ -2,9 +2,7 @@
 from itertools import islice
 from typing import Callable, Dict, Iterable, Optional, Union
 
-from thinc.api import Config, Model
-from thinc.legacy import LegacySequenceCategoricalCrossentropy
-from thinc.types import Floats2d, Ints1d
+from thinc.api import Config, Model, SequenceCategoricalCrossentropy
 
 from ..morphology cimport Morphology
 from ..tokens.doc cimport Doc
@@ -296,7 +294,8 @@ class Morphologizer(Tagger):
         DOCS: https://spacy.io/api/morphologizer#get_loss
         """
         validate_examples(examples, "Morphologizer.get_loss")
-        loss_func = LegacySequenceCategoricalCrossentropy(names=tuple(self.labels), normalize=False)
+        loss_func = SequenceCategoricalCrossentropy(names=self.labels, normalize=False,
+                                                    label_smoothing=self.cfg["label_smoothing"])
         truths = []
         for eg in examples:
             eg_truths = []

spacy/pipeline/senter.pyx

@@ -2,10 +2,7 @@
 from itertools import islice
 from typing import Callable, Dict, Iterable, List, Optional, Union
 
-import srsly
-from thinc.api import Config, Model
-from thinc.legacy import LegacySequenceCategoricalCrossentropy
-from thinc.types import Floats2d, Ints1d
+from thinc.api import Config, Model, SequenceCategoricalCrossentropy
 
 from ..tokens.doc cimport Doc
 
@@ -160,7 +157,7 @@ class SentenceRecognizer(Tagger):
         """
         validate_examples(examples, "SentenceRecognizer.get_loss")
         labels = self.labels
-        loss_func = LegacySequenceCategoricalCrossentropy(names=labels, normalize=False)
+        loss_func = SequenceCategoricalCrossentropy(names=labels, normalize=False)
         truths = []
         for eg in examples:
             eg_truth = []

spacy/pipeline/tagger.pyx

@@ -4,9 +4,7 @@ from itertools import islice
 from typing import Callable, Dict, Iterable, List, Optional, Tuple, Union
 
 import numpy
-import srsly
-from thinc.api import Config, Model, set_dropout_rate
-from thinc.legacy import LegacySequenceCategoricalCrossentropy
+from thinc.api import Config, Model, SequenceCategoricalCrossentropy, set_dropout_rate
 from thinc.types import Floats2d, Ints1d
 
 from ..morphology cimport Morphology
@@ -275,7 +273,7 @@ class Tagger(TrainablePipe):
 
         DOCS: https://spacy.io/api/tagger#get_teacher_student_loss
         """
-        loss_func = LegacySequenceCategoricalCrossentropy(normalize=False)
+        loss_func = SequenceCategoricalCrossentropy(normalize=False)
         d_scores, loss = loss_func(student_scores, teacher_scores)
         if self.model.ops.xp.isnan(loss):
             raise ValueError(Errors.E910.format(name=self.name))
@@ -292,7 +290,12 @@ class Tagger(TrainablePipe):
         DOCS: https://spacy.io/api/tagger#get_loss
         """
         validate_examples(examples, "Tagger.get_loss")
-        loss_func = LegacySequenceCategoricalCrossentropy(names=self.labels, normalize=False, neg_prefix=self.cfg["neg_prefix"])
+        loss_func = SequenceCategoricalCrossentropy(
+            names=self.labels,
+            normalize=False,
+            neg_prefix=self.cfg["neg_prefix"],
+            label_smoothing=self.cfg["label_smoothing"]
+        )
         # Convert empty tag "" to missing value None so that both misaligned
         # tokens and tokens with missing annotation have the default missing
         # value None.

spacy/pipeline/transition_parser.pyx

@@ -22,13 +22,13 @@ from thinc.api import (
     CupyOps,
     NumpyOps,
     Optimizer,
+    SequenceCategoricalCrossentropy,
     chain,
     get_ops,
     set_dropout_rate,
     softmax_activation,
     use_ops,
 )
-from thinc.legacy import LegacySequenceCategoricalCrossentropy
 from thinc.types import Floats2d
 
 from ..ml.parser_model cimport (
@@ -355,7 +355,7 @@ cdef class Parser(TrainablePipe):
 
         DOCS: https://spacy.io/api/dependencyparser#get_teacher_student_loss
         """
-        loss_func = LegacySequenceCategoricalCrossentropy(normalize=False)
+        loss_func = SequenceCategoricalCrossentropy(normalize=False)
         d_scores, loss = loss_func(student_scores, teacher_scores)
         if self.model.ops.xp.isnan(loss):
             raise ValueError(Errors.E910.format(name=self.name))