Lazy Learning Models

This module contains the implementation of the Weighted K-Nearest Neighbors algorithm for lazy learning.

Weighted K-Nearest Neighbors Class

class source.model.lazy.WeightedKNN(alpha, k=5, decay='exponential', bandwidth=1.0, memory_size=100)

Bases: object

Weighted K-Nearest Neighbors with exponential decay weighting.

Parameters:

• alpha (float) – Decay rate for the weights.

• k (int) – Number of nearest neighbors to consider.

• decay (str) – Type of decay to use (“exponential”, “linear” supported).

• bandwidth (float) – Bandwidth for the KDE.

• memory_size (int) – Maximum number of observations to store in memory.

Initialize the WeightedKNN object.

Parameters:

• alpha (float) – Decay rate for the weights.

• k (int) – Number of nearest neighbors to consider.

• decay (str) – Type of decay to use (“exponential”, “linear” supported).

• bandwidth (float) – Bandwidth for the KDE.

• memory_size (int) – Maximum number of observations to store in memory.

__init__(alpha, k=5, decay='exponential', bandwidth=1.0, memory_size=100)

Initialize the WeightedKNN object.

Parameters:

• alpha (float) – Decay rate for the weights.

• k (int) – Number of nearest neighbors to consider.

• decay (str) – Type of decay to use (“exponential”, “linear” supported).

• bandwidth (float) – Bandwidth for the KDE.

• memory_size (int) – Maximum number of observations to store in memory.

lazy_predict(x_query, take='kde')

Predict the value for a new observation using weighted KNN.

Parameters:

• x_query (numpy array) – Query observation vector.

• take (str) – Method to aggregate neighbor labels (“mean” or “weighted_mean” or “kde”).

Returns:

prediction – Predicted value for the query observation.

Return type:

float

update_store(X_new, y_new)

Update the data store with new observations and labels.

Parameters:

• X_new (numpy array) – New observation vectors.

• y_new (numpy array) – New labels corresponding to the observations.

CKDE forecasting based on Weighted K-Nearest Neighbors Model