Tutorial: How to work with custom concepts

While the biaslyze package contains a number of concepts for both English and German, you might want to work with your own concepts. This tutorial will show you how to do that.

First we need to import the Concepts class from the biaslyze package:

from biaslyze.concepts import Concept

Define a custom concept

Now, we can define a new concept. Let's say we want to detect the effect of names in German texts on the prediction of our model. We can define a new concept by passing a List of Dictionaries, where each Dictionary contains a keyword with some metadata.

names_concept = Concept.from_dict_keyword_list(
    name="names",
    lang="de",
    keywords=[{"keyword": "Hans", "function": ["name"]}],
)

Use the custom concept in the CounterfactualBiasDetector

When this is done, we can register the concept with the CounterfactualBiasDetector class. This will make sure that the concept is used when we call the process method of the CounterfactualBiasDetector class.

bias_detector = CounterfactualBiasDetector(lang="de")
bias_detector.register_concept(names_concept)

Now we can finally use the concept by calling the process method of the CounterfactualBiasDetector class. We will use the predict_proba method of the clf object to predict the probabilities of the given texts.

detection_res = bias_detector.process(
    texts=texts,
    predict_func=clf.predict_proba
    concepts_to_consider=["names"],
)