`classification_protocols`

This library provides protocols used in the implementation of machine learning classifier algorithms. Datasets are represented as objects implementing the dataset_protocol protocol. Classifiers are represented as objects implementing the classifier_protocol protocol.

This library also provides reusable shared categories, smoke tests, and test datasets. See below for details.

Logtalk currently provides several classifiers including c45_classifier, knn_classifier, linear_svm_classifier, logistic_regression_classifier, naive_bayes_classifier, nearest_centroid_classifier, and random_forest_classifier. See these libraries documentation for details.

Shared category

The library includes one reusable category intended to be imported by classifier algorithm implementations:

classifier_common — shared diagnostics accessors and classifier export helpers.

This category keeps diagnostics access and file export behavior separate from the algorithm-specific learning, prediction, and pretty-printing code.

Diagnostics

The classifier_common category provides shared accessor predicates such as diagnostics/2, diagnostic/2, and classifier_options/2. These predicates make it possible to inspect learned-classifier metadata without depending on the exact term representation used by a particular classifier implementation.

The detailed contents of the diagnostics data are classifier-dependent. For example, some classifiers report effective training options, while others report structural metadata such as attribute names, feature types, or the number of training examples or models.

Export header format

The shared classifier exporter in the classifier_common category writes a header before the exported clauses in the following format:

% exported classifier predicate: Functor/Arity
% training dataset: Dataset
% dataset prediction schema: Functor(Attribute1, ..., AttributeN, Class)
% diagnostics: Diagnostics
% Functor(Classifier)
Functor(Classifier)

The dataset prediction schema line always uses an ASCII-only title case conversion for the attribute names and class. This line documents the dataset-level prediction interface for readability, even when the exported clauses serialize a model term instead of an executable predictor relation.

When exporting a serialized classifier term, using a noun such as classifier/1 or model/1 is recommended.

API documentation

Open the ../../apis/library_index.html#classification_protocols link in a web browser.

Loading

To load all entities in this library, load the loader.lgt file:

| ?- logtalk_load(classification_protocols(loader)).

Testing

To test this library predicates, shared categories, and datasets, load the tester.lgt file:

| ?- logtalk_load(classification_protocols(tester)).

Test datasets

Several sample datasets are included in the test_datasets directory:

play_tennis.lgt — The classic weather/tennis dataset with 14 examples and 4 discrete attributes (outlook, temperature, humidity, wind). Originally from Quinlan (1986) and widely used in machine learning textbooks including Mitchell (1997). Also available from the UCI Machine Learning Repository: https://archive.ics.uci.edu/dataset/349/tennis+major+tournament+match+statistics
contact_lenses.lgt — A dataset with 24 examples and 4 discrete attributes (age, spectacle prescription, astigmatism, tear production rate) for deciding the type of contact lenses to prescribe. Originally from Cendrowska, J. (1987). PRISM: An algorithm for inducing modular rules. International Journal of Man-Machine Studies, 27(4), 349-370. Available from the UCI Machine Learning Repository: https://archive.ics.uci.edu/dataset/58/lenses
iris.lgt — The classic Iris flower dataset with 150 examples and 4 continuous attributes (sepal length, sepal width, petal length, petal width) for classifying iris species (setosa, versicolor, virginica). Originally from Fisher, R.A. (1936). The use of multiple measurements in taxonomic problems. Annals of Eugenics, 7(2), 179-188. Available from the UCI Machine Learning Repository: https://archive.ics.uci.edu/dataset/53/iris
breast_cancer.lgt — A dataset with 286 examples and 9 discrete attributes (age, menopause, tumor size, inv-nodes, node-caps, degree of malignancy, breast, breast quadrant, irradiation) for predicting breast cancer recurrence events. Contains missing values (9 examples with missing values in the node-caps and breast-quad attributes, represented using anonymous variables). Originally from the Institute of Oncology, University Medical Centre, Ljubljana, Yugoslavia. Donors: Ming Tan and Jeff Schlimmer. Available from the UCI Machine Learning Repository: https://archive.ics.uci.edu/dataset/14/breast+cancer

classification_protocols