cfg parameter is empty when run spacy.load()

[alessandrobokan]

How to reproduce the behaviour

My component class:

class MyComponent(object):
    name = "my_component"

    def __init__(self, nlp, **cfg):
        self.nlp = nlp
        self.categories = cfg.get("categories", "all_categories")  # cfg is EMPTY

I used entry points to add the component my_component to the pipeline. I updated the spacy_factories from setup.py. Then I run python setup.py sdist, generate the lib en_core_web_test_sm, and installed it by pip install.

However, when I run following code:

import spacy

nlp = spacy.load("en_core_web_test_sm", categories=["category1", "category2"])

the parameter cfg from __init__(self, nlp, **cfg): is EMPTY, so I can't get categories.

Why does this happen? Did I do something wrong? Is this a bug? I just followed the documentation.

OBS: **overrides is not passed here https:/explosion/spaCy/blob/master/spacy/util.py#L209

Your Environment

• Operating System: Linux
• Python Version Used: 3.6
• spaCy Version Used: 2.2.3
• Environment Information: virtualenv

[adrianeboyd]

Thanks for the report, this does look like a problem for spacy.load().

@ines may have a better idea, but my first suggestion would be that pipeline_args in overrides should be used to update the config that's passed to the component:

    for name in pipeline:
        if name not in disable:
            config = meta.get("pipeline_args", {}).get(name, {})
            config.update(overrides.get("pipeline_args", {}).get(name, {}))
            factory = factories.get(name, name)
            component = nlp.create_pipe(factory, config=config)
            nlp.add_pipe(component, name=name)

instead of:

spaCy/spacy/util.py

Lines 205 to 210 in 26a90f0

	for name in pipeline:
	if name not in disable:
	config = meta.get("pipeline_args", {}).get(name, {})
	factory = factories.get(name, name)
	component = nlp.create_pipe(factory, config=config)
	nlp.add_pipe(component, name=name)

I think a second possibility would be to store the overrides in nlp somewhere so it's accessible to pipeline components from this point on:

spaCy/spacy/util.py

Line 197 in 26a90f0

nlp = cls(meta=meta, **overrides)

[alessandrobokan]

Hello guys, there is no answer from @adrianeboyd ... I think this issue is kinda important. It appear in the spacy documentation also.

Thanks 👍

[svlandeg]

Addressed by PR #5374, will be fixed in spaCy 2.3 onwards.

[github-actions]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.