Get augmented data

[qeterme]

Hi,
I'm trying to augment my data during training, but sometimes I would like to see the augmented data. How can I do that?

[danieldk]

Thanks for your question! Since you posted this in the Coding & Implementations section, I assume that you'd like to use the augmenters from Python for further exploration. The augmenters themselves are functions that take a Language object and an Example and yield an augmented Example.

The best way to get an augmenter is to retrieve it by its registered name through the spaCy registry. For example, to get the spacy.lower_case.v1, you could use

registry.get("augmenters", "spacy.lower_case.v1")

This returns a function that constructs the augmenter when given the options specified for the augmenter. Since the lower case augmenter only has the level option, we should specify that. E.g.:

augmenter = registry.get("augmenters", "spacy.lower_case.v1")(level=0.5)

augmenter is now a function that you can provide with a Language object and an Example. Here is a small self-contained example:

import spacy
from spacy.util import registry
from spacy.training import Example

# The spaCy registry lets us get the augmenter's builder function. We pass it the
# level argument to construct the augmenter.
augmenter = registry.get("augmenters", "spacy.lower_case.v1")(level=0.5)

nlp = spacy.blank("en")

doc = nlp.make_doc("Let's Augment This Sentence.")
example = Example(doc, doc)

augmented_example = next(augmenter(nlp, example))
# With level=0.5, there is a 50% chance that the example is lowercased.
print(augmented_example.predicted.text)