Choquet-classifier

Implementation of the Choquet classifier. This model was presented in "Learning monotone nonlinear models using the Choquet integral" [1]

Installation

Dependencies

• Python (>=3.9)
• NumPy (>=1.23.3)
• SciPy (>=1.9.1)
• scikit-learn (>=1.1.1)

User installation

• clone this repository or
• Use the package manager pip to install the Choquet classifier.

pip install choquet-classifier-glenscalai

Usage

The application is compatible to scikit-learn and can be used like every other classifier from the scikit-learn library. In order to use the Choquet classifier, import the class ChoquetClassifier from the module choquet_classifier from the package classifier. Some examples are shown below:

Default example

Use the constructor and the function fit to initialize the Choquet classifier for a given dataset.

>>> from choquet_classifier_glenscalai.choquet_classifier import ChoquetClassifier
>>> X = [[1, 3, 2],
...      [1, 0, 3]]
>>> y = [1, 0]
>>> cc = ChoquetClassifier()
>>> cc.fit(X, y)

Use the function predict to classify samples.

>>> Z = [[1, 1, 2],
...      [2, 1, 3]]
>>> cc.predict(Z)
array([0, 0])

Examples with hyper-parameters

>>> from choquet_classifier_glenscalai.choquet_classifier import ChoquetClassifier
>>> X = [[1, 3, 2],
...      [1, 0, 3]]
>>> y = [1, 0]
>>> cc = ChoquetClassifier(additivity=3, regularization=1)
>>> cc.fit(X, y)

Again, the function predict can be used to classify samples. Note the different output compared to the first example.

>>> Z = [[1, 1, 2],
...      [2, 1, 3]]
>>> sc.predict(Z)
array([0, 1])

Example with different class labels

The classes do not have to be labeled with 0 and 1; any integer numbers or strings may be used instead. The smaller label in terms of the relation or lexicographic ordering is given to the negative class, whereas the other label is assigned to the positive class.

The first example contains the class labels 2 and 1. Label 2 is assigned to the positive class and label 1 is assigned to the negative class since 2>1.

>>> from choquet_classifier_glenscalai.choquet_classifier import ChoquetClassifier
>>> X = [[1, 3, 2],
...      [1, 0, 3]]
>>> y = [2, 1]
>>> cc = ChoquetClassifier()
>>> cc.fit(X, y)
>>> Z = [[1, 1, 2],
...      [2, 1, 3]]
>>> sc.predict(Z)
array([1, 1])

The second example contains the class labels 'one' and 'two'. Label 'one' is assigned to the negative class and label 'two' is assigned to the positive class because 'one' comes lexicographically first.

>>> from choquet_classifier_glenscalai.choquet_classifier import ChoquetClassifier
>>> X = [[1, 3, 2],
...      [1, 0, 3]]
>>> y = ['two', 'one']
>>> cc = ChoquetClassifier()
>>> cc.fit(X, y)
>>> Z = [[1, 1, 2],
...      [2, 1, 3]]
>>> sc.predict(Z)
array(['one', 'one'])

Licence

MIT

Reference

[1] Ali Fallah Tehrani, Weiwei Cheng, Krzysztof Dembczynski and Eyke Hüllermeier. Learning monotone nonlinear models using the Choquet integral. 2012.