Issue with creating a Language object from a loaded config.

[Vadim-Maklakov]

Hi all!
I check the training model with CLI using ner_demo project from yours projects folder – model training and predict after loading from HDD very well.
Now I am trying to train the model in Spyder IDE using Language.from_config(config) instead of using CLI.
https://spacy.io/api/language#from_config.
code:

import spacy
from thinc.api import Config
from spacy.language import Language
#Import config setting for train inside Spyder
config = Config().from_disk("/home/mvg/Ewps/Spacy3/config.cfg")
nlp = Language.from_config(config)

output from Spyder console:

ValueError: [E958] Language code defined in config (en) does not match language code of current
Language subclass Language (None). If you want to create an nlp object from a config, make sure 
to use the matching subclass with the language-specific settings and data.

I tried change in [nlp] section on …./site-packages/spacy/default_config.cfg
from lang = null to lang = “en” - but on the other cases I get same ValueError: [E958]

Your Environment

• Operating System: Linux Debuan 10
• Python Version Used: Python 3.8.7
• spaCy Version Used: spaCy version 3.1.2
• Environment Information: config.cfg generated with this command from spacy without any errors
  python -m spacy init config config.cfg --lang en --pipeline transformer,morphologizer,parser,ner,textcat,tagger --optimize accuracy --gpu
  and attachment to this post.
  config.cfg.zip

[Vadim-Maklakov]

Hi all!
Please close this issue and please write somewhere - in the API or doc string for default_config.cfg - that for creating nlp instance from custom config.cfg you must have write those sentences - a) import language b)use name language for creating nlp instance:

from spacy.lang.en import  English
config = Config().from_disk("path_to/config.cfg")
nlp = English.from_config(config)

[Norky101]

Hi @Vadim-Maklakov I am in the same situation as you were. How did you fix this please?

[adrianeboyd]

In general, we strongly recommend using the train CLI because there can be a lot of details to get right in the training loop and it manages a lot of them for you. If you don't want the use the CLI part of the equation, you should still consider using (or at least taking a careful look at) the train_while_improving method from spacy.training.loop.

If you are in doubt when loading manually like this, step through what happens in spacy.load in the code to understand what the required steps are here.