ImportError: cannot import name 'component' from 'spacy.language'

[hgoelen]

Hello,

I installed spaCy through conda on my Windows laptop. Afterwards I downloaded the Japanese corpus ja_core_news_lg (and _sm), but whenever I try to load it, I get the following error message:

---------------------------------------------------------------------------
ImportError                               Traceback (most recent call last)
~\AppData\Local\Temp/ipykernel_3104/42624960.py in <module>
----> 1 nlp = spacy.load('ja_core_news_sm')

~\anaconda3\lib\site-packages\spacy\__init__.py in load(name, vocab, disable, exclude, config)
     49     RETURNS (Language): The loaded nlp object.
     50     """
---> 51     return util.load_model(
     52         name, vocab=vocab, disable=disable, exclude=exclude, config=config
     53     )

~\anaconda3\lib\site-packages\spacy\util.py in load_model(name, vocab, disable, exclude, config)
    345             return get_lang_class(name.replace("blank:", ""))()
    346         if is_package(name):  # installed as package
--> 347             return load_model_from_package(name, **kwargs)
    348         if Path(name).exists():  # path to model data directory
    349             return load_model_from_path(Path(name), **kwargs)

~\anaconda3\lib\site-packages\spacy\util.py in load_model_from_package(name, vocab, disable, exclude, config)
    378     """
    379     cls = importlib.import_module(name)
--> 380     return cls.load(vocab=vocab, disable=disable, exclude=exclude, config=config)
    381 
    382 

~\anaconda3\lib\site-packages\ja_core_news_sm\__init__.py in load(**overrides)
      8 
      9 def load(**overrides):
---> 10     return load_model_from_init_py(__file__, **overrides)

~\anaconda3\lib\site-packages\spacy\util.py in load_model_from_init_py(init_file, vocab, disable, exclude, config)
    538     if not model_path.exists():
    539         raise IOError(Errors.E052.format(path=data_path))
--> 540     return load_model_from_path(
    541         data_path,
    542         vocab=vocab,

~\anaconda3\lib\site-packages\spacy\util.py in load_model_from_path(model_path, meta, vocab, disable, exclude, config)
    413     overrides = dict_to_dot(config)
    414     config = load_config(config_path, overrides=overrides)
--> 415     nlp = load_model_from_config(config, vocab=vocab, disable=disable, exclude=exclude)
    416     return nlp.from_disk(model_path, exclude=exclude, overrides=overrides)
    417 

~\anaconda3\lib\site-packages\spacy\util.py in load_model_from_config(config, vocab, disable, exclude, auto_fill, validate)
    450     # registry, including custom subclasses provided via entry points
    451     lang_cls = get_lang_class(nlp_config["lang"])
--> 452     nlp = lang_cls.from_config(
    453         config,
    454         vocab=vocab,

~\anaconda3\lib\site-packages\spacy\language.py in from_config(cls, config, vocab, disable, exclude, meta, auto_fill, validate)
   1683         # then we would load them twice at runtime: once when we make from config,
   1684         # and then again when we load from disk.
-> 1685         nlp = lang_cls(vocab=vocab, create_tokenizer=create_tokenizer, meta=meta)
   1686         if after_creation is not None:
   1687             nlp = after_creation(nlp)

~\anaconda3\lib\site-packages\spacy\language.py in __init__(self, vocab, max_length, meta, create_tokenizer, batch_size, **kwargs)
    150         # points. The factory decorator applied to these functions takes care
    151         # of the rest.
--> 152         util.registry._entry_point_factories.get_all()
    153 
    154         self._config = DEFAULT_CONFIG.merge(self.default_config)

~\anaconda3\lib\site-packages\catalogue\__init__.py in get_all(self)
    107         result = {}
    108         if self.entry_points:
--> 109             result.update(self.get_entry_points())
    110         for keys, value in REGISTRY.items():
    111             if len(self.namespace) == len(keys) - 1 and all(

~\anaconda3\lib\site-packages\catalogue\__init__.py in get_entry_points(self)
    122         result = {}
    123         for entry_point in AVAILABLE_ENTRY_POINTS.get(self.entry_point_namespace, []):
--> 124             result[entry_point.name] = entry_point.load()
    125         return result
    126 

~\anaconda3\lib\importlib\metadata.py in load(self)
     75         """
     76         match = self.pattern.match(self.value)
---> 77         module = import_module(match.group('module'))
     78         attrs = filter(None, (match.group('attr') or '').split('.'))
     79         return functools.reduce(getattr, attrs, module)

~\anaconda3\lib\importlib\__init__.py in import_module(name, package)
    125                 break
    126             level += 1
--> 127     return _bootstrap._gcd_import(name[level:], package, level)
    128 
    129 

~\anaconda3\lib\importlib\_bootstrap.py in _gcd_import(name, package, level)

~\anaconda3\lib\importlib\_bootstrap.py in _find_and_load(name, import_)

~\anaconda3\lib\importlib\_bootstrap.py in _find_and_load_unlocked(name, import_)

~\anaconda3\lib\importlib\_bootstrap.py in _load_unlocked(spec)

~\anaconda3\lib\importlib\_bootstrap_external.py in exec_module(self, module)

~\anaconda3\lib\importlib\_bootstrap.py in _call_with_frames_removed(f, *args, **kwds)

~\AppData\Roaming\Python\Python38\site-packages\ginza\__init__.py in <module>
      6 from spacy.tokens import Doc, Span, Token
      7 
----> 8 from .bunsetu_recognizer import *
      9 from .compound_splitter import *
     10 from .ene_ontonotes_mapper import ENE_ONTONOTES_MAPPING

~\AppData\Roaming\Python\Python38\site-packages\ginza\bunsetu_recognizer.py in <module>
      1 from typing import Iterable, List
      2 
----> 3 from spacy.language import Language, component
      4 from spacy.tokens import Doc, Span, Token
      5 

ImportError: cannot import name 'component' from 'spacy.language' 

I have tried this many times already, installing spaCy with pip or conda, but nothing seems to fix this. I am using a recent installation of Anaconda; the exact steps I followed to install spaCy are here: amueller/introduction_to_ml_with_python#146

Thanks in advance

[polm]

It looks like you have installed both ginza and spaCy and they are having some weird interaction.

• how did you install the model?
• if you remove ginza, does it work?
• did you install ginza manually?

[polm]

Ginza is a Japanese-specific fork of spaCy. I can tell you have it installed because of this line, for example:

~\AppData\Roaming\Python\Python38\site-packages\ginza\bunsetu_recognizer.py in <module>

It looks like it is installed outside of Anaconda but is being picked up anyway. It should show up in the output of pip freeze. Maybe you can try to pip uninstall ginza?

Another thing to try is using a new conda environment using conda create --name spacy-ja (you can use anything instead of spacy-ja). I think this was resolved a while ago, but we used to sometimes have problems with installs in the default conda env.

[polm]

Oh yeah, another thing, what is the output of spacy info?

[hgoelen]

This worked, thanks! Ginza did show up in pip freeze. Maybe it is because I downloaded the core news corpora?

{'spacy_version': '3.1.3',
 'location': '...',
 'platform': 'Windows-10-10.0.19042-SP0',
 'python_version': '3.8.11',
 'pipelines': {'ja_core_news_lg': '3.1.0'}}

[polm]

Ginza is not part of any official spaCy distribution, and should not be installed with the official Japanese models. It is barely possible we could have accidentally included it somehow, but I just checked in a blank env to verify that's not the case. I suspect you must have installed it in some other way.

I also tried intentionally installing spaCy and ginza in the same environment and loading the Japanese model, and that worked without issue as expected. I've never heard of spaCy and ginza interacting like they did with your error, I'm honestly not sure what could cause that.

[hgoelen]

I don't know what went wrong either, but thanks for helping me out!