Update TextCatBOW to use the fixed SparseLinear layer

spacy/cli/templates/quickstart_training.jinja

@@ -271,8 +271,9 @@ grad_factor = 1.0
 @layers = "reduce_mean.v1"
 
 [components.textcat.model.linear_model]
-@architectures = "spacy.TextCatBOW.v2"
+@architectures = "spacy.TextCatBOW.v3"
 exclusive_classes = true
+length = 262144
 ngram_size = 1
 no_output_layer = false
 
@@ -308,8 +309,9 @@ grad_factor = 1.0
 @layers = "reduce_mean.v1"
 
 [components.textcat_multilabel.model.linear_model]
-@architectures = "spacy.TextCatBOW.v2"
+@architectures = "spacy.TextCatBOW.v3"
 exclusive_classes = false
+length = 262144
 ngram_size = 1
 no_output_layer = false
 
@@ -542,14 +544,15 @@ nO = null
 width = ${components.tok2vec.model.encode.width}
 
 [components.textcat.model.linear_model]
-@architectures = "spacy.TextCatBOW.v2"
+@architectures = "spacy.TextCatBOW.v3"
 exclusive_classes = true
+length = 262144
 ngram_size = 1
 no_output_layer = false
 
 {% else -%}
 [components.textcat.model]
-@architectures = "spacy.TextCatBOW.v2"
+@architectures = "spacy.TextCatBOW.v3"
 exclusive_classes = true
 ngram_size = 1
 no_output_layer = false
@@ -570,15 +573,17 @@ nO = null
 width = ${components.tok2vec.model.encode.width}
 
 [components.textcat_multilabel.model.linear_model]
-@architectures = "spacy.TextCatBOW.v2"
+@architectures = "spacy.TextCatBOW.v3"
 exclusive_classes = false
+length = 262144
 ngram_size = 1
 no_output_layer = false
 
 {% else -%}
 [components.textcat_multilabel.model]
-@architectures = "spacy.TextCatBOW.v2"
+@architectures = "spacy.TextCatBOW.v3"
 exclusive_classes = false
+length = 262144
 ngram_size = 1
 no_output_layer = false
 {%- endif %}

spacy/errors.py

@@ -983,6 +983,7 @@ class Errors(metaclass=ErrorsWithCodes):
              "predicted docs when training {component}.")
     E1055 = ("The 'replace_listener' callback expects {num_params} parameters, "
              "but only callbacks with one or three parameters are supported")
+    E1056 = ("The `TextCatBOW` architecture expects a length of at least 1, was {length}.")
 
 
 # Deprecated model shortcuts, only used in errors and warnings

spacy/ml/models/textcat.py

@@ -1,5 +1,5 @@
 from functools import partial
-from typing import List, Optional, cast
+from typing import List, Optional, Tuple, cast
 
 from thinc.api import (
     Dropout,
@@ -12,6 +12,7 @@
     Relu,
     Softmax,
     SparseLinear,
+    SparseLinear_v2,
     chain,
     clone,
     concatenate,
@@ -25,9 +26,10 @@
 )
 from thinc.layers.chain import init as init_chain
 from thinc.layers.resizable import resize_linear_weighted, resize_model
-from thinc.types import Floats2d
+from thinc.types import ArrayXd, Floats2d
 
 from ...attrs import ORTH
+from ...errors import Errors
 from ...tokens import Doc
 from ...util import registry
 from ..extract_ngrams import extract_ngrams
@@ -95,10 +97,48 @@ def build_bow_text_classifier(
     ngram_size: int,
     no_output_layer: bool,
     nO: Optional[int] = None,
+) -> Model[List[Doc], Floats2d]:
+    return _build_bow_text_classifier(
+        exclusive_classes=exclusive_classes,
+        ngram_size=ngram_size,
+        no_output_layer=no_output_layer,
+        nO=nO,
+        sparse_linear=SparseLinear(nO=nO),
+    )
+
+
+@registry.architectures("spacy.TextCatBOW.v3")
+def build_bow_text_classifier_v3(
+    exclusive_classes: bool,
+    ngram_size: int,
+    no_output_layer: bool,
+    length: int = 262144,
+    nO: Optional[int] = None,
+) -> Model[List[Doc], Floats2d]:
+    if length < 1:
+        raise ValueError(Errors.E1056.format(length=length))
+
+    # Find k such that 2**(k-1) < length <= 2**k.
+    length = 2 ** (length - 1).bit_length()
+
+    return _build_bow_text_classifier(
+        exclusive_classes=exclusive_classes,
+        ngram_size=ngram_size,
+        no_output_layer=no_output_layer,
+        nO=nO,
+        sparse_linear=SparseLinear_v2(nO=nO, length=length),
+    )
+
+
+def _build_bow_text_classifier(
+    exclusive_classes: bool,
+    ngram_size: int,
+    no_output_layer: bool,
+    sparse_linear: Model[Tuple[ArrayXd, ArrayXd, ArrayXd], ArrayXd],
+    nO: Optional[int] = None,
 ) -> Model[List[Doc], Floats2d]:
     fill_defaults = {"b": 0, "W": 0}
     with Model.define_operators({">>": chain}):
-        sparse_linear = SparseLinear(nO=nO)
         output_layer = None
         if not no_output_layer:
             fill_defaults["b"] = NEG_VALUE

spacy/pipeline/textcat.py

@@ -36,17 +36,19 @@
 depth = 2
 
 [model.linear_model]
-@architectures = "spacy.TextCatBOW.v2"
+@architectures = "spacy.TextCatBOW.v3"
 exclusive_classes = true
+length = 262144
 ngram_size = 1
 no_output_layer = false
 """
 DEFAULT_SINGLE_TEXTCAT_MODEL = Config().from_str(single_label_default_config)["model"]
 
 single_label_bow_config = """
 [model]
-@architectures = "spacy.TextCatBOW.v2"
+@architectures = "spacy.TextCatBOW.v3"
 exclusive_classes = true
+length = 262144
 ngram_size = 1
 no_output_layer = false
 """

spacy/pipeline/textcat_multilabel.py

@@ -35,16 +35,17 @@
 depth = 2
 
 [model.linear_model]
-@architectures = "spacy.TextCatBOW.v2"
+@architectures = "spacy.TextCatBOW.v3"
 exclusive_classes = false
+length = 262144
 ngram_size = 1
 no_output_layer = false
 """
 DEFAULT_MULTI_TEXTCAT_MODEL = Config().from_str(multi_label_default_config)["model"]
 
 multi_label_bow_config = """
 [model]
-@architectures = "spacy.TextCatBOW.v2"
+@architectures = "spacy.TextCatBOW.v3"
 exclusive_classes = false
 ngram_size = 1
 no_output_layer = false

spacy/tests/pipeline/test_pipe_factories.py

@@ -203,7 +203,7 @@ def test_pipe_class_component_model():
             "@architectures": "spacy.TextCatEnsemble.v2",
             "tok2vec": DEFAULT_TOK2VEC_MODEL,
             "linear_model": {
-                "@architectures": "spacy.TextCatBOW.v2",
+                "@architectures": "spacy.TextCatBOW.v3",
                 "exclusive_classes": False,
                 "ngram_size": 1,
                 "no_output_layer": False,

spacy/tests/pipeline/test_textcat.py

@@ -414,7 +414,7 @@ def test_implicit_label(name, get_examples):
 @pytest.mark.parametrize(
     "name,textcat_config",
     [
-        # BOW
+        # BOW V1
         ("textcat", {"@architectures": "spacy.TextCatBOW.v1", "exclusive_classes": True, "no_output_layer": False, "ngram_size": 3}),
         ("textcat", {"@architectures": "spacy.TextCatBOW.v1", "exclusive_classes": True, "no_output_layer": True, "ngram_size": 3}),
         ("textcat_multilabel", {"@architectures": "spacy.TextCatBOW.v1", "exclusive_classes": False, "no_output_layer": False, "ngram_size": 3}),
@@ -451,11 +451,11 @@ def test_no_resize(name, textcat_config):
 @pytest.mark.parametrize(
     "name,textcat_config",
     [
-        # BOW
-        ("textcat", {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": True, "no_output_layer": False, "ngram_size": 3}),
-        ("textcat", {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": True, "no_output_layer": True, "ngram_size": 3}),
-        ("textcat_multilabel", {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": False, "no_output_layer": False, "ngram_size": 3}),
-        ("textcat_multilabel", {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": False, "no_output_layer": True, "ngram_size": 3}),
+        # BOW V3
+        ("textcat", {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": True, "no_output_layer": False, "ngram_size": 3}),
+        ("textcat", {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": True, "no_output_layer": True, "ngram_size": 3}),
+        ("textcat_multilabel", {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": False, "no_output_layer": False, "ngram_size": 3}),
+        ("textcat_multilabel", {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": False, "no_output_layer": True, "ngram_size": 3}),
         # CNN
         ("textcat", {"@architectures": "spacy.TextCatCNN.v2", "tok2vec": DEFAULT_TOK2VEC_MODEL, "exclusive_classes": True}),
         ("textcat_multilabel", {"@architectures": "spacy.TextCatCNN.v2", "tok2vec": DEFAULT_TOK2VEC_MODEL, "exclusive_classes": False}),
@@ -480,11 +480,11 @@ def test_resize(name, textcat_config):
 @pytest.mark.parametrize(
     "name,textcat_config",
     [
-        # BOW
-        ("textcat", {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": True, "no_output_layer": False, "ngram_size": 3}),
-        ("textcat", {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": True, "no_output_layer": True, "ngram_size": 3}),
-        ("textcat_multilabel", {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": False, "no_output_layer": False, "ngram_size": 3}),
-        ("textcat_multilabel", {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": False, "no_output_layer": True, "ngram_size": 3}),
+        # BOW v3
+        ("textcat", {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": True, "no_output_layer": False, "ngram_size": 3}),
+        ("textcat", {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": True, "no_output_layer": True, "ngram_size": 3}),
+        ("textcat_multilabel", {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": False, "no_output_layer": False, "ngram_size": 3}),
+        ("textcat_multilabel", {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": False, "no_output_layer": True, "ngram_size": 3}),
         # CNN
         ("textcat", {"@architectures": "spacy.TextCatCNN.v2", "tok2vec": DEFAULT_TOK2VEC_MODEL, "exclusive_classes": True}),
         ("textcat_multilabel", {"@architectures": "spacy.TextCatCNN.v2", "tok2vec": DEFAULT_TOK2VEC_MODEL, "exclusive_classes": False}),
@@ -693,9 +693,14 @@ def test_overfitting_IO_multi():
         ("textcat", TRAIN_DATA_SINGLE_LABEL, {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": True, "ngram_size": 4, "no_output_layer": False}),
         ("textcat_multilabel", TRAIN_DATA_MULTI_LABEL, {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": False, "ngram_size": 3, "no_output_layer": True}),
         ("textcat", TRAIN_DATA_SINGLE_LABEL, {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": True, "ngram_size": 2, "no_output_layer": True}),
+        # BOW V3
+        ("textcat_multilabel", TRAIN_DATA_MULTI_LABEL, {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": False, "ngram_size": 1, "no_output_layer": False}),
+        ("textcat", TRAIN_DATA_SINGLE_LABEL, {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": True, "ngram_size": 4, "no_output_layer": False}),
+        ("textcat_multilabel", TRAIN_DATA_MULTI_LABEL, {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": False, "ngram_size": 3, "no_output_layer": True}),
+        ("textcat", TRAIN_DATA_SINGLE_LABEL, {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": True, "ngram_size": 2, "no_output_layer": True}),
         # ENSEMBLE V2
-        ("textcat_multilabel", TRAIN_DATA_MULTI_LABEL, {"@architectures": "spacy.TextCatEnsemble.v2", "tok2vec": DEFAULT_TOK2VEC_MODEL, "linear_model": {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": False, "ngram_size": 1, "no_output_layer": False}}),
-        ("textcat", TRAIN_DATA_SINGLE_LABEL, {"@architectures": "spacy.TextCatEnsemble.v2", "tok2vec": DEFAULT_TOK2VEC_MODEL, "linear_model": {"@architectures": "spacy.TextCatBOW.v2", "exclusive_classes": True, "ngram_size": 5, "no_output_layer": False}}),
+        ("textcat_multilabel", TRAIN_DATA_MULTI_LABEL, {"@architectures": "spacy.TextCatEnsemble.v2", "tok2vec": DEFAULT_TOK2VEC_MODEL, "linear_model": {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": False, "ngram_size": 1, "no_output_layer": False}}),
+        ("textcat", TRAIN_DATA_SINGLE_LABEL, {"@architectures": "spacy.TextCatEnsemble.v2", "tok2vec": DEFAULT_TOK2VEC_MODEL, "linear_model": {"@architectures": "spacy.TextCatBOW.v3", "exclusive_classes": True, "ngram_size": 5, "no_output_layer": False}}),
         # CNN V2
         ("textcat", TRAIN_DATA_SINGLE_LABEL, {"@architectures": "spacy.TextCatCNN.v2", "tok2vec": DEFAULT_TOK2VEC_MODEL, "exclusive_classes": True}),
         ("textcat_multilabel", TRAIN_DATA_MULTI_LABEL, {"@architectures": "spacy.TextCatCNN.v2", "tok2vec": DEFAULT_TOK2VEC_MODEL, "exclusive_classes": False}),

spacy/tests/test_cli_app.py

@@ -238,7 +238,7 @@ def test_project_push_pull(project_dir):
 
 def test_find_function_valid():
     # example of architecture in main code base
-    function = "spacy.TextCatBOW.v2"
+    function = "spacy.TextCatBOW.v3"
     result = CliRunner().invoke(app, ["find-function", function, "-r", "architectures"])
     assert f"Found registered function '{function}'" in result.stdout
     assert "textcat.py" in result.stdout
@@ -257,7 +257,7 @@ def test_find_function_valid():
 
 def test_find_function_invalid():
     # invalid registry
-    function = "spacy.TextCatBOW.v2"
+    function = "spacy.TextCatBOW.v3"
     registry = "foobar"
     result = CliRunner().invoke(
         app, ["find-function", function, "--registry", registry]

spacy/tests/test_misc.py

@@ -376,8 +376,9 @@ def test_util_dot_section():
     factory = "textcat"
 
     [components.textcat.model]
-    @architectures = "spacy.TextCatBOW.v2"
+    @architectures = "spacy.TextCatBOW.v3"
     exclusive_classes = true
+    length = 262144
     ngram_size = 1
     no_output_layer = false
     """