Augmenty for spacy training pipe config.cfg

[leonballonigomes]

Good afternoon,
Ive spent sometime studying the package and I would like to know if its possible to use it with the config.cfg file proposed by Spacy https://spacy.io/usage/training.
If yes, could you send me some info about how to do it, please?

Thank you.

[KennethEnevoldsen]

Indeed it is!

I have an example here where I use it to train DaCy (a danish pipeline for spacy):

[corpora.train]
@readers = "spacy.Corpus.v1"
path = ${paths:train}
max_length = 500
gold_preproc = false
limit = 0

[corpora.train.augmenter]
@augmenters = "dacy_augmenter.v1"

Where dacy_augmenter.v1 can be any augmenter in augmenty (mine is a custom one I made in this script).

When calling spacy train using your new config.cfg it will not know that the augmenty augmenters exist. The way to resolve this is to use the --code argument and pass it a python file (let us call it augmenty_augmenters.py) this file can look like so:

import augmenty
# will add all augmenters to spacy catalogue

So you final command will look something like this:
spacy train [...] --code augmenty_augmenters.py

But you might want to use multiple of the augmenters and can easily combine these (see here for an example). You can then add your newly constructed augmenter using the spacy decorator: @spacy.registry.augmenters("my_cool_augmenter.v1")

I also wrote an example of this in this issue 😄

[leonballonigomes]

Thank you! I'll try this.