Struct MeanShift

pub struct MeanShift { /* private fields */ }

Mean Shift clustering algorithm implementation.

Mean Shift is a centroid-based clustering algorithm that works by iteratively shifting data points towards areas of higher density. Each data point moves in the direction of the mean of points within its current window until convergence. The algorithm does not require specifying the number of clusters in advance.

Fields

• bandwidth - The kernel bandwidth parameter that determines the search radius. Larger values lead to fewer clusters.
• max_iter - Maximum number of iterations to prevent infinite loops.
• tol - Convergence tolerance threshold. Points are considered converged when they move less than this value.
• bin_seeding - Whether to use bin seeding strategy for faster algorithm execution.
• cluster_all - Whether to assign all points to clusters, including potential noise.

Examples

use rustyml::machine_learning::meanshift::MeanShift;
use ndarray::Array2;

// Create a 2D dataset
let data = Array2::<f64>::from_shape_vec((10, 2),
    vec![1.0, 2.0, 1.1, 2.2, 0.9, 1.9, 1.0, 2.1,
         10.0, 10.0, 10.2, 9.9, 10.1, 10.0, 9.9, 9.8,
         5.0, 5.0, 5.1, 4.9]).unwrap();

// Create a MeanShift instance with default parameters
let mut ms = MeanShift::default();

// Fit the model and predict cluster labels
let labels = ms.fit_predict(&data);

// Get the cluster centers
let centers = ms.get_cluster_centers().unwrap();

Notes

• If unsure about an appropriate bandwidth value, use the estimate_bandwidth function.
• The bandwidth parameter significantly affects algorithm performance and should be chosen carefully based on data characteristics.
• For large datasets, setting bin_seeding = true can improve performance.

Implementations

impl MeanShift

pub fn new( bandwidth: f64, max_iter: Option<usize>, tol: Option<f64>, bin_seeding: Option<bool>, cluster_all: Option<bool>, ) -> Self

Creates a new MeanShift instance with the specified parameters.

Parameters

• bandwidth - The bandwidth parameter that determines the size of the kernel.
• max_iter - The maximum number of iterations for the mean shift algorithm.
• tol - The convergence threshold for the algorithm.
• bin_seeding - Whether to use bin seeding for initialization.
• cluster_all - Whether to assign all points to clusters, even those far from any centroid.

Returns

• Self - A new MeanShift instance.

pub fn get_cluster_centers(&self) -> Result<Array2<f64>, ModelError>

Gets the cluster centers found by the algorithm.

Returns

• Ok(Array2<f64>) - A Result containing the cluster centers as a ndarray Array2<f64>

• Err(ModelError::NotFitted) - If the model has not been fitted yet

pub fn get_labels(&self) -> Result<Array1<usize>, ModelError>

Gets the cluster labels assigned to each data point.

Returns

• Ok(Array1<usize>) - A Result containing the cluster labels as a ndarray Array1<usize>
• Err(ModelError::NotFitted) - If the model has not been fitted yet

pub fn get_n_iter(&self) -> Result<usize, ModelError>

Gets the number of iterations the algorithm performed.

Returns

• Ok(usize) - A Result containing the number of iterations or an error
• Err(ModelError::NotFitted) - If the model has not been fitted yet

pub fn get_n_samples_per_center(&self) -> Result<Array1<usize>, ModelError>

Gets the number of samples per cluster center.

Returns

• Ok(Array1<usize>) - A Result containing the number of samples per center as a ndarray Array1<usize>
• Err(ModelError::NotFitted) - If the model has not been fitted yet

pub fn get_bandwidth(&self) -> f64

Gets the bandwidth parameter value.

Returns

• f64 - The bandwidth value.

pub fn get_max_iter(&self) -> usize

Gets the maximum number of iterations.

Returns

• usize - The maximum number of iterations.

pub fn get_tol(&self) -> f64

Gets the convergence tolerance.

Returns

• f64 - The tolerance value.

pub fn get_bin_seeding(&self) -> bool

Gets the bin seeding setting.

Returns

• bool - A boolean indicating whether bin seeding is enabled.

pub fn get_cluster_all(&self) -> bool

Gets the cluster_all setting.

Returns

• bool - A boolean indicating whether all points are assigned to clusters.

pub fn fit(&mut self, x: &Array2<f64>) -> Result<&mut Self, ModelError>

Fits the MeanShift clustering model to the input data.

Parameters

• x - The input data as a ndarray Array2<f64> where each row is a sample.

Returns

• Ok(&mut Self) - A mutable reference to the fitted model
• Err(ModelError::InputValidationError(&str)) - Input does not match expectation

pub fn predict(&self, x: &Array2<f64>) -> Result<Array1<usize>, ModelError>

Predicts cluster labels for the input data.

Parameters

• x - The input data as a ndarray Array2<f64> where each row is a sample.

Returns

• Ok(Array1<usize>) - containing the predicted cluster labels.
• Err(ModelError::NotFitted) - If the model has not been fitted yet

pub fn fit_predict( &mut self, x: &Array2<f64>, ) -> Result<Array1<usize>, ModelError>

Fits the model to the input data and predicts cluster labels.

Parameters

• x - The input data as a ndarray Array2<f64> where each row is a sample.

Returns

• Ok(Array1<usize>) - containing the predicted cluster labels.
• Err(ModelError::InputValidationError(&str)) - Input does not match expectation

Trait Implementations

impl Clone for MeanShift

fn clone(&self) -> MeanShift

Returns a copy of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for MeanShift

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for MeanShift

fn default() -> Self

Returns the “default value” for a type.