Why doesn't the Textcat have a tok2vec?

[taghouti-ghofrane]

Hello all , i am confused in generating base_config.cfg from spacy documentation!
when i select components textcat , hardware:cpu , optimize for : efficiency ,
the pipeline has just Textcat:

[nlp]
lang = "en"
pipeline = ["textcat"]
batch_size = 1000

Is it Normal ?

How we can Train a model without Tok2vec in The pipeline ?
I trained the model just with that Pipeline , it didnt give good result !

• Python Version Used: 3.9.0
• spaCy Version Used: 3.2

[adrianeboyd]

The simplest textcat model uses a bag of words only, so it doesn't require a tok2vec, but as you noticed, the performance is not always as good as some of the other (less efficient) options. In that case, you can try other architectures instead, like spacy.TextCatEnsemble.v2:

https://spacy.io/api/architectures#textcat