ValueError: [E150] when using customized model

[tobias-hahn]

Hi,

I am working on a NER project and would kindly ask for your help.

For this project, named entities in text snippets taken from invoices have to be recognized, such as the invoice's date or id. This works quite well, but the default tokenizer from the "de" module (as well as the "en" one) does not isolate all tokens as required. There are two cases so far, exemplified by the following code:

>>> import spacy
>>> nlp = spacy.blank("de")
>>> text = "123 456(foo   ean:12345677"
>>> for token in nlp(text):
...     print(token)
... 
123
456(foo
  
ean:12345677

What I need is to isolate an entity even if a matching closing parenthesis is missing or if two entities are separated by a colon, like this:

123
456
(
foo
  
ean
:
12345677

Since this project is supposed to work with the project / config workflow of SpaCy 3, I set about to create custom_code.py, which looks like this:

import spacy
from spacy.lang.de import German

class CustomGermanDefaults(German.Defaults):
    infixes = ("\(",) + ("\(",) + (":",) + tuple(German.Defaults.infixes)

@spacy.registry.languages("custom_de")
class CustomGerman(German):
    lang = "custom_de"
    Defaults = CustomGermanDefaults

This file works during the conversion of my labeled data to SpaCy's binary format with conversion.py...

import srsly
import typer
import warnings
from pathlib import Path
import os
import spacy
from spacy.tokens import DocBin

import sys
import inspect
currentdir = os.path.dirname(os.path.abspath(inspect.getfile(inspect.currentframe())))
parentdir = os.path.dirname(currentdir)
sys.path.insert(0, parentdir)
import custom_code as cc

def convert(input_path: Path, output_path: Path):
    nlp = spacy.blank("custom_de")
    db = DocBin()
    for text, annot in srsly.read_json(input_path):
        doc = nlp.make_doc(text)
        ents = []
        for start, end, label in annot["entities"]:
            span = doc.char_span(start, end, label=label)
            if span is None:
                msg = f"Skipping entity [{start}, {end}, {label}] in the following text because the character span '{doc.text[start:end]}' does not align with token boundaries:\n\n{repr(text)}\n"
                warnings.warn(msg)
            else:
                ents.append(span)
        doc.ents = ents
        db.add(doc)
    Path(os.path.split(output_path)[0]).mkdir(exist_ok=True)
    db.to_disk(output_path)

if __name__ == "__main__":
    typer.run(convert)

...but when I incorporate into the SpaCy workflow, I encounter this error message:

  File "/usr/local/anaconda3/envs/spacy/lib/python3.8/site-packages/spacy/language.py", line 179, in __init__
    raise ValueError(Errors.E150.format(nlp=self.lang, vocab=vocab.lang))
ValueError: [E150] The language of the `nlp` object and the `vocab` should be the same, but found 'de' and 'custom_de' respectively.

The config.cfglooks like this:

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

[system]
gpu_allocator = null
seed = 0

[nlp]
lang = "custom_de"
pipeline = ["tok2vec","ner"]
batch_size = 1000
disabled = []
before_creation = null
after_creation = null
after_pipeline_creation = null
tokenizer = {"@tokenizers":"spacy.Tokenizer.v1"}

[components]

[components.ner]
factory = "ner"
incorrect_spans_key = null
moves = null
scorer = {"@scorers":"spacy.ner_scorer.v1"}
update_with_oracle_cut_size = 100

[components.ner.model]
@architectures = "spacy.TransitionBasedParser.v2"
state_type = "ner"
extra_state_tokens = false
hidden_width = 64
maxout_pieces = 2
use_upper = true
nO = null

[components.ner.model.tok2vec]
@architectures = "spacy.Tok2VecListener.v1"
width = ${components.tok2vec.model.encode.width}
upstream = "*"

[components.tok2vec]
factory = "tok2vec"

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

[components.tok2vec.model.embed]
@architectures = "spacy.MultiHashEmbed.v2"
width = ${components.tok2vec.model.encode.width}
attrs = ["ORTH","SHAPE"]
rows = [5000,2500]
include_static_vectors = true

[components.tok2vec.model.encode]
@architectures = "spacy.MaxoutWindowEncoder.v2"
width = 256
depth = 8
window_size = 1
maxout_pieces = 3

[corpora]

[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 = []
annotating_components = []
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 = false

[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]
ents_f = 1.0
ents_p = 0.0
ents_r = 0.0
ents_per_type = null

[pretraining]

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

[initialize.components]

[initialize.tokenizer]      

And the cli command, which throws the error is this:

python3 -m spacy debug data --verbose configs/${vars.name}/config.cfg --paths.train corpus/${vars.name}/train.spacy --paths.dev corpus/${vars.name}/dev.spacy --code custom_code.py

The default "de" model is referenced nowhere in my code, so I am surprised to see it mentioned in the error message.

Do you have any idea what is happening here? If you do, I would be grateful to hear from you.

Best regards.

[adrianeboyd]

I think the default de is coming from the vectors in de_core_news_lg. I'll take a look at what's going on...

As an alternative, you can customize the tokenizer for de / German as described here instead:

https://spacy.io/usage/training#custom-tokenizer

It's simpler overall, also when packaging the pipeline later, if you don't have a custom language. Just for modifying the tokenizer settings, it shouldn't be necessary to have a custom language.

[tobias-hahn]

I think the default de is coming from the vectors in de_core_news_lg. I'll take a look at what's going on...

As an alternative, you can customize the tokenizer for de / German as described here instead:

https://spacy.io/usage/training#custom-tokenizer

It's simpler overall, also when packaging the pipeline later, if you don't have a custom language. Just for modifying the tokenizer settings, it shouldn't be necessary to have a custom language.

Thank you for you quick reply. My code is running now with your recommended solution, which truly is way simpler.