enable argument for spacy.load()

[rmitsch]

Goals

Introduce enable argument to spacy.load(), behaving analogously to and disable.

Description

enable takes precedence over their negative counterparts. I.e.: if enable is set, disable has to be consistent. Otherwise an error is raised. Internally, enable is resolved to disable by disabling those components in the pipeline that are not part of include.

Example:

spacy.load("en_core_web_sm", enable=["tagger", "parser"])

This disables all components other than the tagger and the parser.

Types of change

New feature, new tests.

Checklist

• I confirm that I have the right to submit this contribution under the project's MIT license.
• I ran the tests, and all new and existing tests passed.
• My changes don't require a change to the documentation, or if they do, I've added all required information.

[svlandeg]

I'd love to hear some thoughts/reviews by other colleagues as well, but in principle I think this will be really useful. It makes the code & docs slightly more complex, but it will have a huge impact on many use-cases where you really just want one or 2 components and disable everything else. It feels like such a common use-case that we might as well cater for it...

[rmitsch]

The Python38Mac tests fail due to a config mismatch (the new enabled property seems to be missing). IMO that shouldn't be the case though, and the other jobs in the build matrix succeed...ideas? 🤔

[adrianeboyd]

The Python 3.8 mac build includes some additional CLI tests that aren't run in the other builds (just to avoid excessive downloads/time), so these failures do matter, and aren't particular to python 3.8 / macos.

I didn't realize that the plan was to add this setting to the config? I thought this was just adding new keyword arguments to spacy.load?

Edited to add: I don't think it makes sense to have two conflicting features in the config like disable and include. Someone writing a config can figure out what the right settings are with pipeline and disable. Someone loading a model with spacy.load has a good reason to want an include option that makes it easier to selectively load things.

[rmitsch]

The Python 3.8 mac build includes some additional CLI tests that aren't run in the other builds (just to avoid excessive downloads/time), so these failures do matter, and aren't particular to python 3.8 / macos.

I didn't realize that the plan was to add this setting to the config? I thought this was just adding new keyword arguments to spacy.load?

Edited to add: I don't think it makes sense to have two conflicting features in the config like disable and include. Someone writing a config can figure out what the right settings are with pipeline and disable. Someone loading a model with spacy.load has a good reason to want an include option that makes it easier to selectively load things.

Adding it to the config isn't necessary. It was my assumption we'd want to handle enable the same way as disable, which would mean adding it to the config. I think you have a point there, I'd be ok with removing it from the config again. What does @svlandeg think?

[svlandeg]

Agreed with Adriane's points!

[rmitsch]

Removed enabled from the config. This is the last remaining open question from my side.

[rmitsch]

Removed include. Updated website docs.

[polm]

I'm kind of late here, but I agree this would be a great feature to have - for short bits of code this can be clearer and more succint than disable.

[svlandeg]

Looks good to me! Just had a few more minor comments but should otherwise be good to merge :-)

[svlandeg]

Nice work, Raphael!