Support custom token/lexeme attribute for vectors (#12625)

* Support custom token/lexeme attribute for vectors

* Fix imports

* Back off to ORTH without Vectors.attr

* Fallback if vectors.attr doesn't exist

* Update docs

spacy/cli/init_pipeline.py

@@ -32,6 +32,7 @@ def init_vectors_cli(
     name: Optional[str] = Opt(None, "--name", "-n", help="Optional name for the word vectors, e.g. en_core_web_lg.vectors"),
     verbose: bool = Opt(False, "--verbose", "-V", "-VV", help="Display more information for debugging purposes"),
     jsonl_loc: Optional[Path] = Opt(None, "--lexemes-jsonl", "-j", help="Location of JSONL-formatted attributes file", hidden=True),
+    attr: str = Opt("ORTH", "--attr", "-a", help="Optional token attribute to use for vectors, e.g. LOWER or NORM"),
     # fmt: on
 ):
     """Convert word vectors for use with spaCy. Will export an nlp object that
@@ -50,6 +51,7 @@ def init_vectors_cli(
         prune=prune,
         name=name,
         mode=mode,
+        attr=attr,
     )
     msg.good(f"Successfully converted {len(nlp.vocab.vectors)} vectors")
     nlp.to_disk(output_dir)

spacy/errors.py

@@ -216,6 +216,9 @@ class Warnings(metaclass=ErrorsWithCodes):
     W123 = ("Argument `enable` with value {enable} does not contain all values specified in the config option "
             "`enabled` ({enabled}). Be aware that this might affect other components in your pipeline.")
     W124 = ("{host}:{port} is already in use, using the nearest available port {serve_port} as an alternative.")
+    W125 = ("The StaticVectors key_attr is no longer used. To set a custom "
+            "key attribute for vectors, configure it through Vectors(attr=) or "
+            "'spacy init vectors --attr'")
 
 
 class Errors(metaclass=ErrorsWithCodes):

spacy/ml/staticvectors.py

@@ -1,3 +1,4 @@
+import warnings
 from typing import Callable, List, Optional, Sequence, Tuple, cast
 
 from thinc.api import Model, Ops, registry
@@ -5,7 +6,8 @@ from thinc.initializers import glorot_uniform_init
 from thinc.types import Floats1d, Floats2d, Ints1d, Ragged
 from thinc.util import partial
 
-from ..errors import Errors
+from ..attrs import ORTH
+from ..errors import Errors, Warnings
 from ..tokens import Doc
 from ..vectors import Mode
 from ..vocab import Vocab
@@ -24,6 +26,8 @@ def StaticVectors(
     linear projection to control the dimensionality. If a dropout rate is
     specified, the dropout is applied per dimension over the whole batch.
     """
+    if key_attr != "ORTH":
+        warnings.warn(Warnings.W125, DeprecationWarning)
     return Model(
         "static_vectors",
         forward,
@@ -40,9 +44,9 @@ def forward(
     token_count = sum(len(doc) for doc in docs)
     if not token_count:
         return _handle_empty(model.ops, model.get_dim("nO"))
-    key_attr: int = model.attrs["key_attr"]
-    keys = model.ops.flatten([cast(Ints1d, doc.to_array(key_attr)) for doc in docs])
     vocab: Vocab = docs[0].vocab
+    key_attr: int = getattr(vocab.vectors, "attr", ORTH)
+    keys = model.ops.flatten([cast(Ints1d, doc.to_array(key_attr)) for doc in docs])
     W = cast(Floats2d, model.ops.as_contig(model.get_param("W")))
     if vocab.vectors.mode == Mode.default:
         V = model.ops.asarray(vocab.vectors.data)

spacy/tests/vocab_vectors/test_vectors.py

@@ -402,6 +402,7 @@ def test_vectors_serialize():
         row_r = v_r.add("D", vector=OPS.asarray([10, 20, 30, 40], dtype="f"))
         assert row == row_r
         assert_equal(OPS.to_numpy(v.data), OPS.to_numpy(v_r.data))
+        assert v.attr == v_r.attr
 
 
 def test_vector_is_oov():
@@ -646,3 +647,32 @@ def test_equality():
     vectors1.resize((5, 9))
     vectors2.resize((5, 9))
     assert vectors1 == vectors2
+
+
+def test_vectors_attr():
+    data = numpy.asarray([[0, 0, 0], [1, 2, 3], [9, 8, 7]], dtype="f")
+    # default ORTH
+    nlp = English()
+    nlp.vocab.vectors = Vectors(data=data, keys=["A", "B", "C"])
+    assert nlp.vocab.strings["A"] in nlp.vocab.vectors.key2row
+    assert nlp.vocab.strings["a"] not in nlp.vocab.vectors.key2row
+    assert nlp.vocab["A"].has_vector is True
+    assert nlp.vocab["a"].has_vector is False
+    assert nlp("A")[0].has_vector is True
+    assert nlp("a")[0].has_vector is False
+
+    # custom LOWER
+    nlp = English()
+    nlp.vocab.vectors = Vectors(data=data, keys=["a", "b", "c"], attr="LOWER")
+    assert nlp.vocab.strings["A"] not in nlp.vocab.vectors.key2row
+    assert nlp.vocab.strings["a"] in nlp.vocab.vectors.key2row
+    assert nlp.vocab["A"].has_vector is True
+    assert nlp.vocab["a"].has_vector is True
+    assert nlp("A")[0].has_vector is True
+    assert nlp("a")[0].has_vector is True
+    # add a new vectors entry
+    assert nlp.vocab["D"].has_vector is False
+    assert nlp.vocab["d"].has_vector is False
+    nlp.vocab.set_vector("D", numpy.asarray([4, 5, 6]))
+    assert nlp.vocab["D"].has_vector is True
+    assert nlp.vocab["d"].has_vector is True

spacy/tokens/doc.pyx

@@ -35,6 +35,7 @@ from ..attrs cimport (
     LENGTH,
     MORPH,
     NORM,
+    ORTH,
     POS,
     SENT_START,
     SPACY,
@@ -613,13 +614,26 @@ cdef class Doc:
         """
         if "similarity" in self.user_hooks:
             return self.user_hooks["similarity"](self, other)
-        if isinstance(other, (Lexeme, Token)) and self.length == 1:
+        attr = getattr(self.vocab.vectors, "attr", ORTH)
-            if self.c[0].lex.orth == other.orth:
+        cdef Token this_token
+        cdef Token other_token
+        cdef Lexeme other_lex
+        if len(self) == 1 and isinstance(other, Token):
+            this_token = self[0]
+            other_token = other
+            if Token.get_struct_attr(this_token.c, attr) == Token.get_struct_attr(other_token.c, attr):
                 return 1.0
-        elif isinstance(other, (Span, Doc)) and len(self) == len(other):
+        elif len(self) == 1 and isinstance(other, Lexeme):
+            this_token = self[0]
+            other_lex = other
+            if Token.get_struct_attr(this_token.c, attr) == Lexeme.get_struct_attr(other_lex.c, attr):
+                return 1.0
+        elif isinstance(other, (Doc, Span)) and len(self) == len(other):
             similar = True
-            for i in range(self.length):
+            for i in range(len(self)):
-                if self[i].orth != other[i].orth:
+                this_token = self[i]
+                other_token = other[i]
+                if Token.get_struct_attr(this_token.c, attr) != Token.get_struct_attr(other_token.c, attr):
                     similar = False
                     break
             if similar:

spacy/tokens/span.pyx

@@ -8,13 +8,14 @@ import numpy
 from thinc.api import get_array_module
 
 from ..attrs cimport *
-from ..attrs cimport attr_id_t
+from ..attrs cimport ORTH, attr_id_t
 from ..lexeme cimport Lexeme
 from ..parts_of_speech cimport univ_pos_t
 from ..structs cimport LexemeC, TokenC
 from ..symbols cimport dep
 from ..typedefs cimport attr_t, flags_t, hash_t
 from .doc cimport _get_lca_matrix, get_token_attr, token_by_end, token_by_start
+from .token cimport Token
 
 from ..errors import Errors, Warnings
 from ..util import normalize_slice
@@ -341,13 +342,26 @@ cdef class Span:
         """
         if "similarity" in self.doc.user_span_hooks:
             return self.doc.user_span_hooks["similarity"](self, other)
-        if len(self) == 1 and hasattr(other, "orth"):
+        attr = getattr(self.doc.vocab.vectors, "attr", ORTH)
-            if self[0].orth == other.orth:
+        cdef Token this_token
+        cdef Token other_token
+        cdef Lexeme other_lex
+        if len(self) == 1 and isinstance(other, Token):
+            this_token = self[0]
+            other_token = other
+            if Token.get_struct_attr(this_token.c, attr) == Token.get_struct_attr(other_token.c, attr):
+                return 1.0
+        elif len(self) == 1 and isinstance(other, Lexeme):
+            this_token = self[0]
+            other_lex = other
+            if Token.get_struct_attr(this_token.c, attr) == Lexeme.get_struct_attr(other_lex.c, attr):
                 return 1.0
         elif isinstance(other, (Doc, Span)) and len(self) == len(other):
             similar = True
             for i in range(len(self)):
-                if self[i].orth != getattr(other[i], "orth", None):
+                this_token = self[i]
+                other_token = other[i]
+                if Token.get_struct_attr(this_token.c, attr) != Token.get_struct_attr(other_token.c, attr):
                     similar = False
                     break
             if similar:

spacy/tokens/token.pyx

@@ -28,6 +28,7 @@ from ..attrs cimport (
     LIKE_EMAIL,
     LIKE_NUM,
     LIKE_URL,
+    ORTH,
 )
 from ..lexeme cimport Lexeme
 from ..symbols cimport conj
@@ -214,11 +215,17 @@ cdef class Token:
         """
         if "similarity" in self.doc.user_token_hooks:
             return self.doc.user_token_hooks["similarity"](self, other)
-        if hasattr(other, "__len__") and len(other) == 1 and hasattr(other, "__getitem__"):
+        attr = getattr(self.doc.vocab.vectors, "attr", ORTH)
-            if self.c.lex.orth == getattr(other[0], "orth", None):
+        cdef Token this_token = self
+        cdef Token other_token
+        cdef Lexeme other_lex
+        if isinstance(other, Token):
+            other_token = other
+            if Token.get_struct_attr(this_token.c, attr) == Token.get_struct_attr(other_token.c, attr):
                 return 1.0
-        elif hasattr(other, "orth"):
+        elif isinstance(other, Lexeme):
-            if self.c.lex.orth == other.orth:
+            other_lex = other
+            if Token.get_struct_attr(this_token.c, attr) == Lexeme.get_struct_attr(other_lex.c, attr):
                 return 1.0
         if self.vocab.vectors.n_keys == 0:
             warnings.warn(Warnings.W007.format(obj="Token"))
@@ -415,7 +422,7 @@ cdef class Token:
             return self.doc.user_token_hooks["has_vector"](self)
         if self.vocab.vectors.size == 0 and self.doc.tensor.size != 0:
             return True
-        return self.vocab.has_vector(self.c.lex.orth)
+        return self.vocab.has_vector(Token.get_struct_attr(self.c, self.vocab.vectors.attr))
 
     @property
     def vector(self):
@@ -431,7 +438,7 @@ cdef class Token:
         if self.vocab.vectors.size == 0 and self.doc.tensor.size != 0:
             return self.doc.tensor[self.i]
         else:
-            return self.vocab.get_vector(self.c.lex.orth)
+            return self.vocab.get_vector(Token.get_struct_attr(self.c, self.vocab.vectors.attr))
 
     @property
     def vector_norm(self):

spacy/training/initialize.py

@@ -216,9 +216,14 @@ def convert_vectors(
     prune: int,
     name: Optional[str] = None,
     mode: str = VectorsMode.default,
+    attr: str = "ORTH",
 ) -> None:
     vectors_loc = ensure_path(vectors_loc)
     if vectors_loc and vectors_loc.parts[-1].endswith(".npz"):
+        if attr != "ORTH":
+            raise ValueError(
+                "ORTH is the only attribute supported for vectors in .npz format."
+            )
         nlp.vocab.vectors = Vectors(
             strings=nlp.vocab.strings, data=numpy.load(vectors_loc.open("rb"))
         )
@@ -246,11 +251,15 @@ def convert_vectors(
                 nlp.vocab.vectors = Vectors(
                     strings=nlp.vocab.strings,
                     data=vectors_data,
+                    attr=attr,
                     **floret_settings,
                 )
             else:
                 nlp.vocab.vectors = Vectors(
-                    strings=nlp.vocab.strings, data=vectors_data, keys=vector_keys
+                    strings=nlp.vocab.strings,
+                    data=vectors_data,
+                    keys=vector_keys,
+                    attr=attr,
                 )
                 nlp.vocab.deduplicate_vectors()
     if name is None:

spacy/vectors.pyx

@@ -15,9 +15,11 @@ from thinc.api import Ops, get_array_module, get_current_ops
 from thinc.backends import get_array_ops
 from thinc.types import Floats2d
 
+from .attrs cimport ORTH, attr_id_t
 from .strings cimport StringStore
 
 from . import util
+from .attrs import IDS
 from .errors import Errors, Warnings
 from .strings import get_string_id
 
@@ -64,8 +66,9 @@ cdef class Vectors:
     cdef readonly uint32_t hash_seed
     cdef readonly unicode bow
     cdef readonly unicode eow
+    cdef readonly attr_id_t attr
 
-    def __init__(self, *, strings=None, shape=None, data=None, keys=None, name=None, mode=Mode.default, minn=0, maxn=0, hash_count=1, hash_seed=0, bow="<", eow=">"):
+    def __init__(self, *, strings=None, shape=None, data=None, keys=None, name=None, mode=Mode.default, minn=0, maxn=0, hash_count=1, hash_seed=0, bow="<", eow=">", attr="ORTH"):
         """Create a new vector store.
 
         strings (StringStore): The string store.
@@ -80,6 +83,8 @@ cdef class Vectors:
         hash_seed (int): The floret hash seed (default: 0).
         bow (str): The floret BOW string (default: "<").
         eow (str): The floret EOW string (default: ">").
+        attr (Union[int, str]): The token attribute for the vector keys
+            (default: "ORTH").
 
         DOCS: https://spacy.io/api/vectors#init
         """
@@ -103,6 +108,14 @@ cdef class Vectors:
         self.hash_seed = hash_seed
         self.bow = bow
         self.eow = eow
+        if isinstance(attr, (int, long)):
+            self.attr = attr
+        else:
+            attr = attr.upper()
+            if attr == "TEXT":
+                attr = "ORTH"
+            self.attr = IDS.get(attr, ORTH)
+
         if self.mode == Mode.default:
             if data is None:
                 if shape is None:
@@ -546,6 +559,7 @@ cdef class Vectors:
                 "hash_seed": self.hash_seed,
                 "bow": self.bow,
                 "eow": self.eow,
+                "attr": self.attr,
             }
 
     def _set_cfg(self, cfg):
@@ -556,6 +570,7 @@ cdef class Vectors:
         self.hash_seed = cfg.get("hash_seed", 0)
         self.bow = cfg.get("bow", "<")
         self.eow = cfg.get("eow", ">")
+        self.attr = cfg.get("attr", ORTH)
 
     def to_disk(self, path, *, exclude=tuple()):
         """Save the current state to a directory.

spacy/vocab.pyx

@@ -365,8 +365,13 @@ cdef class Vocab:
             self[orth]
         # Make prob negative so it sorts by rank ascending
         # (key2row contains the rank)
-        priority = [(-lex.prob, self.vectors.key2row[lex.orth], lex.orth)
+        priority = []
-                    for lex in self if lex.orth in self.vectors.key2row]
+        cdef Lexeme lex
+        cdef attr_t value
+        for lex in self:
+            value = Lexeme.get_struct_attr(lex.c, self.vectors.attr)
+            if value in self.vectors.key2row:
+                priority.append((-lex.prob, self.vectors.key2row[value], value))
         priority.sort()
         indices = xp.asarray([i for (prob, i, key) in priority], dtype="uint64")
         keys = xp.asarray([key for (prob, i, key) in priority], dtype="uint64")
@@ -399,8 +404,10 @@ cdef class Vocab:
         """
         if isinstance(orth, str):
             orth = self.strings.add(orth)
-        if self.has_vector(orth):
+        cdef Lexeme lex = self[orth]
-            return self.vectors[orth]
+        key = Lexeme.get_struct_attr(lex.c, self.vectors.attr)
+        if self.has_vector(key):
+            return self.vectors[key]
         xp = get_array_module(self.vectors.data)
         vectors = xp.zeros((self.vectors_length,), dtype="f")
         return vectors
@@ -416,15 +423,16 @@ cdef class Vocab:
         """
         if isinstance(orth, str):
             orth = self.strings.add(orth)
-        if self.vectors.is_full and orth not in self.vectors:
+        cdef Lexeme lex = self[orth]
+        key = Lexeme.get_struct_attr(lex.c, self.vectors.attr)
+        if self.vectors.is_full and key not in self.vectors:
             new_rows = max(100, int(self.vectors.shape[0]*1.3))
             if self.vectors.shape[1] == 0:
                 width = vector.size
             else:
                 width = self.vectors.shape[1]
             self.vectors.resize((new_rows, width))
-        lex = self[orth]  # Add word to vocab if necessary
+        row = self.vectors.add(key, vector=vector)
-        row = self.vectors.add(orth, vector=vector)
         if row >= 0:
             lex.rank = row
 
@@ -439,7 +447,9 @@ cdef class Vocab:
         """
         if isinstance(orth, str):
             orth = self.strings.add(orth)
-        return orth in self.vectors
+        cdef Lexeme lex = self[orth]
+        key = Lexeme.get_struct_attr(lex.c, self.vectors.attr)
+        return key in self.vectors
 
     property lookups:
         def __get__(self):

website/docs/api/architectures.mdx

@@ -303,7 +303,7 @@ mapped to a zero vector. See the documentation on
 | `nM`        | The width of the static vectors. ~~Optional[int]~~                                                                                                                                                                      |
 | `dropout`   | Optional dropout rate. If set, it's applied per dimension over the whole batch. Defaults to `None`. ~~Optional[float]~~                                                                                                 |
 | `init_W`    | The [initialization function](https://thinc.ai/docs/api-initializers). Defaults to [`glorot_uniform_init`](https://thinc.ai/docs/api-initializers#glorot_uniform_init). ~~Callable[[Ops, Tuple[int, ...]]], FloatsXd]~~ |
-| `key_attr`  | Defaults to `"ORTH"`. ~~str~~                                                                                                                                                                                           |
+| `key_attr`  | This setting is ignored in spaCy v3.6+. To set a custom key attribute for vectors, configure it through [`Vectors`](/api/vectors) or [`spacy init vectors`](/api/cli#init-vectors). Defaults to `"ORTH"`. ~~str~~       |
 | **CREATES** | The model using the architecture. ~~Model[List[Doc], Ragged]~~                                                                                                                                                          |
 
 ### spacy.FeatureExtractor.v1 {id="FeatureExtractor"}

website/docs/api/cli.mdx

@@ -211,7 +211,8 @@ $ python -m spacy init vectors [lang] [vectors_loc] [output_dir] [--prune] [--tr
 | `output_dir`       | Pipeline output directory. Will be created if it doesn't exist. ~~Path (positional)~~                                                                                                                                                                               |
 | `--truncate`, `-t` | Number of vectors to truncate to when reading in vectors file. Defaults to `0` for no truncation. ~~int (option)~~                                                                                                                                                  |
 | `--prune`, `-p`    | Number of vectors to prune the vocabulary to. Defaults to `-1` for no pruning. ~~int (option)~~                                                                                                                                                                     |
-| `--mode`, `-m`     | Vectors mode: `default` or [`floret`](https://github.com/explosion/floret). Defaults to `default`. ~~Optional[str] \(option)~~                                                                                                                                      |
+| `--mode`, `-m`     | Vectors mode: `default` or [`floret`](https://github.com/explosion/floret). Defaults to `default`. ~~str \(option)~~                                                                                                                                                |
+| `--attr`, `-a`     | Token attribute to use for vectors, e.g. `LOWER` or `NORM`) Defaults to `ORTH`. ~~str \(option)~~                                                                                                                                                                   |
 | `--name`, `-n`     | Name to assign to the word vectors in the `meta.json`, e.g. `en_core_web_md.vectors`. ~~Optional[str] \(option)~~                                                                                                                                                   |
 | `--verbose`, `-V`  | Print additional information and explanations. ~~bool (flag)~~                                                                                                                                                                                                      |
 | `--help`, `-h`     | Show help message and available arguments. ~~bool (flag)~~                                                                                                                                                                                                          |

website/docs/api/vectors.mdx

@@ -60,6 +60,7 @@ modified later.
 | `hash_seed` <Tag variant="new">3.2</Tag>  | The floret hash seed (default: `0`). ~~int~~                                                                                                                                           |
 | `bow` <Tag variant="new">3.2</Tag>        | The floret BOW string (default: `"<"`). ~~str~~                                                                                                                                        |
 | `eow` <Tag variant="new">3.2</Tag>        | The floret EOW string (default: `">"`). ~~str~~                                                                                                                                        |
+| `attr` <Tag variant="new">3.6</Tag>       | The token attribute for the vector keys (default: `"ORTH"`). ~~Union[int, str]~~                                                                                                       |
 
 ## Vectors.\_\_getitem\_\_ {id="getitem",tag="method"}
 
@@ -454,7 +455,8 @@ Load state from a binary string.
 ## Attributes {id="attributes"}
 
 | Name                                | Description                                                                                                                                                          |
-| --------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
+| ----------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | `data`                              | Stored vectors data. `numpy` is used for CPU vectors, `cupy` for GPU vectors. ~~Union[numpy.ndarray[ndim=1, dtype=float32], cupy.ndarray[ndim=1, dtype=float32]]~~   |
 | `key2row`                           | Dictionary mapping word hashes to rows in the `Vectors.data` table. ~~Dict[int, int]~~                                                                               |
 | `keys`                              | Array keeping the keys in order, such that `keys[vectors.key2row[key]] == key`. ~~Union[numpy.ndarray[ndim=1, dtype=float32], cupy.ndarray[ndim=1, dtype=float32]]~~ |
+| `attr` <Tag variant="new">3.6</Tag> | The token attribute for the vector keys. ~~int~~                                                                                                                     |