`knn_regression`

k-Nearest Neighbors regressor supporting continuous and mixed-feature datasets. Learns lazily by storing encoded training rows and predicts targets as the weighted average of the k nearest neighbors.

The library implements the regressor_protocol defined in the regression_protocols library, learns lazily by storing encoded training rows, and predicts numeric targets using the weighted average of the nearest encoded neighbors.

API documentation

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

Loading

To load this library, load the loader.lgt file:

| ?- logtalk_load(knn_regression(loader)).

Testing

To test this library predicates, load the tester.lgt file:

| ?- logtalk_load(knn_regression(tester)).

To run the performance benchmark suite, load the tester_performance.lgt file:

| ?- logtalk_load(knn_regression(tester_performance)).

Features

Distance-Based Regression: Predicts targets using the weighted average of the nearest neighbors.
Multiple Metrics: Supports Euclidean, Manhattan, Chebyshev, and Minkowski distance metrics over encoded numeric feature vectors.
Weighting Schemes: Supports uniform, inverse-distance, and Gaussian weighting of neighbors.
Continuous and Mixed Features: Supports continuous attributes and categorical attributes encoded using reference-level dummy coding derived from the declared dataset attribute values.
Optional Feature Scaling: Continuous attributes can be standardized using z-score scaling.
Missing Values: Missing numeric and categorical values represented using anonymous variables are encoded using explicit missing-value indicator features.
Diagnostics Metadata: Learned regressors record model name, target, training example count, encoded feature count, and effective options, accessible using the shared regression diagnostics predicates.
Model Export: Learned regressors can be exported as predicate clauses or written to a file.
Reference Benchmarks: Includes a dedicated performance suite reporting training time, RMSE, and MAE for representative regression datasets.

Regressor representation

The learned regressor is represented by default as:

knn_regressor(Encoders, Rows, Diagnostics)

The exported predicate clauses therefore use the shape:

Functor(Encoders, Rows, Diagnostics)

In this representation, Rows stores encoded feature vectors paired with numeric targets and Diagnostics stores training metadata including the effective options.

Diagnostics syntax

The diagnostics/2 predicate returns a list of metadata terms with the form:

[
    model(knn_regression),
    target(Target),
    training_example_count(TrainingExampleCount),
    options(Options),
    encoded_feature_count(FeatureCount)
]

Where:

model(knn_regression) identifies the learning algorithm that produced the regressor.
target(Target) stores the target attribute name declared by the training dataset.
training_example_count(TrainingExampleCount) stores the number of examples used during training.
options(Options) stores the effective learning options after merging the user options with the library defaults.
encoded_feature_count(FeatureCount) stores the number of numeric features induced by the encoder list, including missing-value indicator features.

Use the regression_protocols diagnostic/2 and regressor_options/2 helper predicates when you only need a single metadata term or the effective options.

Options

The learn/3 predicate accepts the following options:

k/1: Number of nearest neighbors considered for each prediction. Smaller values make predictions more local; larger values smooth them by averaging over more training rows. The default is 3.
distance_metric/1: Distance function used to compare encoded feature vectors. Accepted values are euclidean, manhattan, chebyshev, and minkowski. The default is euclidean.
weight_scheme/1: Neighbor weighting policy used when averaging targets. Accepted values are uniform, distance, and gaussian. The default is uniform.
minkowski_power/1: Exponent used when distance_metric(minkowski) is selected. Larger values increase the influence of larger coordinate differences. The default is 3.0.
feature_scaling/1: Controls z-score standardization of continuous attributes before storing rows and encoding prediction requests. Accepted values are true and false. The default is true.

knn_regression