Error: Could not find function in registry 'misc' after packaging a successfully trained model with code

[adamkgoldfarb]

Hello!

I have been happily training a spancat component with a custom suggester, but after packaging up a model-best I am unable to load the model with spacy. My code setup closely mimics this snippet.

Output from spacy info:

spaCy version    3.2.2                         
Location         /Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/spacy
Platform         macOS-12.2.1-x86_64-i386-64bit
Python version   3.9.10                        
Pipelines        en_core_web_sm (3.2.0), en_core_web_trf (3.2.0), en_core_web_lg (3.2.0)

My config:

[paths]
train = null
dev = null
vectors = "spacy_floret_vectors"
init_tok2vec = null

[system]
gpu_allocator = "pytorch"
seed = 0

[nlp]
lang = "en"
pipeline = ["tok2vec", "tagger", "parser", "senter", "attribute_ruler", "spancat"]
batch_size = 1000
disabled = []
before_creation = null
after_creation = null
after_pipeline_creation = null
tokenizer = {"@tokenizers":"spacy.Tokenizer.v1"}

[components]

[components.tok2vec]
source = "en_core_web_sm"

[components.tagger]
source = "en_core_web_sm"

[components.parser]
source = "en_core_web_sm"

[components.senter]
source = "en_core_web_sm"

[components.attribute_ruler]
source = "en_core_web_sm"

[components.spancat]
factory = "spancat"
max_positive = null
scorer = {"@scorers":"spacy.spancat_scorer.v1"}
spans_key = "sc"
threshold = 0.5

[components.spancat.model]
@architectures = "spacy.SpanCategorizer.v1"

[components.spancat.model.reducer]
@layers = "spacy.mean_max_reducer.v1"
hidden_size = 128

[components.spancat.model.scorer]
@layers = "spacy.LinearLogistic.v1"
nO = null
nI = null

[components.spancat.model.tok2vec]
@architectures = "spacy.Tok2Vec.v2"

[components.spancat.model.tok2vec.embed]
@architectures = "spacy.MultiHashEmbed.v2"
width = 128
rows = [5000,2500,2500,1000,1000,2500,2500]
attrs = ["ORTH","NORM","SHAPE","PREFIX","SUFFIX","POS","DEP"]
include_static_vectors = true

[components.spancat.model.tok2vec.encode]
@architectures = "spacy.TorchBiLSTMEncoder.v1"
width = 128
depth = 2
dropout = ${training.dropout}

[components.spancat.suggester]
@misc = "sent_clause_suggester.v1"
# @misc = "spacy.ngram_range_suggester.v1"
# min_size = 1
# max_size = 100

[corpora]

[corpora.pretrain]
@readers = "spacy.Corpus.v1"
path = ${paths.raw_text}
max_length = 0
gold_preproc = false
limit = 0
augmenter = null

[corpora.dev]
@readers = "spacy.Corpus.v1"
path = ${paths.dev}
max_length = 0
gold_preproc = false
limit = 0
augmenter = null

[corpora.train]
@readers = "spacy.Corpus.v1"
path = ${paths.train}
max_length = 0
gold_preproc = false
limit = 0
augmenter = null

[training]
dev_corpus = "corpora.dev"
train_corpus = "corpora.train"
seed = ${system.seed}
gpu_allocator = ${system.gpu_allocator}
dropout = 0.1
accumulate_gradient = 1
patience = 1600
max_epochs = 0
max_steps = 20000
eval_frequency = 200
frozen_components = ["tok2vec", "tagger", "parser", "senter", "attribute_ruler"]
annotating_components = ["tok2vec", "tagger", "parser", "senter", "attribute_ruler"]
before_to_disk = null

[training.batcher]
@batchers = "spacy.batch_by_words.v1"
discard_oversize = false
tolerance = 0.2
get_length = null

[training.batcher.size]
@schedules = "compounding.v1"
start = 100
stop = 1000
compound = 1.001
t = 0.0

[training.logger]
@loggers = "spacy.ConsoleLogger.v1"
progress_bar = true

[training.optimizer]
@optimizers = "Adam.v1"
beta1 = 0.9
beta2 = 0.999
L2_is_weight_decay = true
L2 = 0.01
grad_clip = 1.0
use_averages = false
eps = 0.00000001
learn_rate = 0.001

[training.score_weights]
spans_sc_f = 0
spans_sc_p = 0.4
spans_sc_r = 0.6
tag_acc = null
dep_uas = null
dep_las = null
sents_p = null
sents_r = null
sents_f = null

[pretraining]
max_epochs = 1000
dropout = 0.2
n_save_every = null
n_save_epoch = null
component = "spancat"
layer = "tok2vec"
corpus = "corpora.pretrain"

[pretraining.batcher]
@batchers = "spacy.batch_by_words.v1"
size = 3000
discard_oversize = false
tolerance = 0.2
get_length = null

[pretraining.objective]
@architectures = "spacy.PretrainCharacters.v1"
maxout_pieces = 3
hidden_size = 300
n_characters = 4

[pretraining.optimizer]
@optimizers = "Adam.v1"
beta1 = 0.9
beta2 = 0.999
L2_is_weight_decay = true
L2 = 0.01
grad_clip = 1.0
use_averages = true
eps = 0.00000001
learn_rate = 0.001

[initialize]
vectors = ${paths.vectors}
init_tok2vec = ${paths.init_tok2vec}
vocab_data = null
lookups = null
before_init = null
after_init = null

[initialize.components]

[initialize.components.spancat]

[initialize.components.spancat.labels]
@readers = "spacy.read_labels.v1"
path = "spancat.json"

[initialize.tokenizer]

And the full stacktrace:

>> spacy.load("packages/en_feedback_par_lg_27_test-0.0.1/en_feedback_par_lg_27_test/en_feedback_par_lg_27_test-0.0.1")
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/spacy/__init__.py", line 51, in load
    return util.load_model(
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/spacy/util.py", line 422, in load_model
    return load_model_from_path(Path(name), **kwargs)  # type: ignore[arg-type]
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/spacy/util.py", line 488, in load_model_from_path
    nlp = load_model_from_config(config, vocab=vocab, disable=disable, exclude=exclude)
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/spacy/util.py", line 525, in load_model_from_config
    nlp = lang_cls.from_config(
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/spacy/language.py", line 1782, in from_config
    nlp.add_pipe(
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/spacy/language.py", line 792, in add_pipe
    pipe_component = self.create_pipe(
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/spacy/language.py", line 674, in create_pipe
    resolved = registry.resolve(cfg, validate=validate)
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/thinc/config.py", line 729, in resolve
    resolved, _ = cls._make(
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/thinc/config.py", line 778, in _make
    filled, _, resolved = cls._fill(
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/thinc/config.py", line 833, in _fill
    filled[key], validation[v_key], final[key] = cls._fill(
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/thinc/config.py", line 832, in _fill
    promise_schema = cls.make_promise_schema(value, resolve=resolve)
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/thinc/config.py", line 1023, in make_promise_schema
    func = cls.get(reg_name, func_name)
  File "/Users/adamgoldfarb/anaconda3/envs/boop_env/lib/python3.9/site-packages/spacy/util.py", line 141, in get
    raise RegistryError(
catalogue.RegistryError: [E893] Could not find function 'sent_clause_suggester.v1' in function registry 'misc'. If you're using a custom function, make sure the code is available. If the function is provided by a third-party package, e.g. spacy-transformers, make sure the package is installed in your environment.

Available names: spacy.CandidateGenerator.v1, spacy.EmptyKB.v1, spacy.KBFromFile.v1, spacy.LookupsDataLoader.v1, spacy.ngram_range_suggester.v1, spacy.ngram_suggester.v1

Finally, the code for the suggester itself, which is in a spancat_suggester.py file passed to --code during training and packaging.

from typing import Optional, Iterable, cast
from thinc.api import get_current_ops, Ops
from thinc.types import Ragged, Ints1d

import spacy

from spacy.matcher import Matcher
from spacy.pipeline.spancat import Suggester
from spacy.tokens import Doc
from spacy.util import registry


@registry.misc("sent_clause_suggester.v1")
def build_sent_clause_suggester() -> Suggester:
    def sent_clause_suggester(
        docs: Iterable[Doc], *, ops: Optional[Ops] = None
    ) -> Ragged:

        nlp = spacy.load("en_core_web_sm", exclude=["lemmatizer", "ner"])
        # nlp = spacy.blank("en")
        # nlp.add_pipe("sentencizer")
        clause_matcher = Matcher(nlp.vocab)
        noun_pattern = [
            {"POS": {"IN": ["NOUN"]}},
            {"POS": {"IN": ["PRON", "SCONJ", "CCONJ"]}},
        ]

        punct_pattern = [
            {"POS": {"IN": ["PUNCT"]}},
            {
                "POS": {
                    "IN": [
                        "PRON",
                        "NOUN",
                        "DET",
                        "VERB",
                        "CCONJ",
                        "SCONJ",
                        "ADV",
                        "ADP",
                    ]
                }
            },
        ]

        cconj_pattern = [
            {"POS": {"IN": ["CCONJ"]}},
            {"POS": {"IN": ["PRON", "VERB", "NOUN", "DET", "AUX", "CCONJ", "SCONJ"]}},
        ]

        sconj_pattern = [
            {"POS": {"IN": ["SCONJ"]}},
            {"POS": {"IN": ["PRON", "VERB", "NOUN", "DET", "AUX", "CCONJ", "SCONJ"]}},
        ]

        # clause_matcher.add("clause_noun", [noun_pattern])
        clause_matcher.add("punctuation", [punct_pattern])
        clause_matcher.add("conjunction", [cconj_pattern])
        clause_matcher.add("subordinating", [sconj_pattern])

        if ops is None:
            ops = get_current_ops()
        spans = []
        lengths = []
        for doc in docs:
            doc = nlp(doc)
            if len(list(doc.sents)) >= 1:
                sent_suggestions = [[sent.start, sent.end] for sent in doc.sents]
                clause_suggestions = []
                for sent in doc.sents:
                    matches = clause_matcher(sent)
                    for match_id, start, end in matches:
                        string_id = nlp.vocab.strings[match_id]
                        if string_id == "punctuation":
                            clause_suggestions.append([start + 1, sent.end])
                        elif string_id == "conjuction":
                            clause_suggestions.append([start + 1, sent.end])
                        elif string_id == "subordinating":
                            clause_suggestions.append([start, sent.end])
                suggestions = sent_suggestions
                for clause in clause_suggestions:
                    if clause not in suggestions:
                        suggestions.append(clause)
                suggestions = ops.asarray(suggestions, dtype="i")
                if suggestions.shape[0] > 0:
                    spans.append(suggestions)
                    lengths.append(suggestions.shape[0])
                else:
                    lengths.append(0)
            else:
                lengths.append(0)
        lengths_array = cast(Ints1d, ops.asarray(lengths, dtype="i"))
        if len(spans) > 0:
            output = Ragged(ops.xp.vstack(spans), lengths_array)
        else:
            output = Ragged(ops.xp.zeros((0, 0), dtype="i"), lengths_array)

        assert output.dataXd.ndim == 2
        return output

    return sent_clause_suggester

This is the first time I've packaged a model rather than used the output directory, so not sure if the directory structure for the package looks right:

packages/en_feedback_par_lg_27_test-0.0.1
├── MANIFEST.in
├── README.md
├── dist
│   └── en_feedback_par_lg_27_test-0.0.1.tar.gz
├── en_feedback_par_lg_27_test
│   ├── __init__.py
│   ├── en_feedback_par_lg_27_test-0.0.1
│   │   ├── README.md
│   │   ├── attribute_ruler
│   │   │   └── patterns
│   │   ├── config.cfg
│   │   ├── meta.json
│   │   ├── parser
│   │   │   ├── cfg
│   │   │   ├── model
│   │   │   └── moves
│   │   ├── senter
│   │   │   ├── cfg
│   │   │   └── model
│   │   ├── spancat
│   │   │   ├── cfg
│   │   │   └── model
│   │   ├── tagger
│   │   │   ├── cfg
│   │   │   └── model
│   │   ├── tok2vec
│   │   │   ├── cfg
│   │   │   └── model
│   │   ├── tokenizer
│   │   └── vocab
│   │       ├── key2row
│   │       ├── lookups.bin
│   │       ├── strings.json
│   │       ├── vectors
│   │       └── vectors.cfg
│   ├── meta.json
│   └── spancat_suggester.py
├── en_feedback_par_lg_27_test.egg-info
│   ├── PKG-INFO
│   ├── SOURCES.txt
│   ├── dependency_links.txt
│   ├── entry_points.txt
│   ├── not-zip-safe
│   ├── requires.txt
│   └── top_level.txt
├── meta.json
└── setup.py

Thanks in advance for any guidance you can provide! Appreciate the efforts of the whole team.

Adam

[adrianeboyd]

The code is only loaded automatically from the installed package, not from the directory. You need to install the model package from dist/ (either .tar.gz or .whl depending on the --build options) and then load it using the package name, just like en_core_web_sm.

[adamkgoldfarb]

Thanks! That should have been obvious. I will bashfully return to the command line...