Struct ForgetfulKernelMatrix 

pub struct ForgetfulKernelMatrix { /* private fields */ }

Kernel matrix with exponential forgetting for concept drift adaptation.

Older samples are weighted by λ^age, allowing the model to adapt to changing data distributions.

Example

use tensorlogic_sklears_kernels::{ForgetfulKernelMatrix, ForgetfulConfig, RbfKernel, RbfKernelConfig, Kernel};

let kernel = RbfKernel::new(RbfKernelConfig::new(0.5)).unwrap();
let config = ForgetfulConfig::with_lambda(0.95).unwrap();
let mut forgetful = ForgetfulKernelMatrix::new(Box::new(kernel), config);

// Add samples (older ones get downweighted)
forgetful.add_sample(vec![1.0, 2.0]).unwrap();
forgetful.add_sample(vec![3.0, 4.0]).unwrap();
forgetful.add_sample(vec![5.0, 6.0]).unwrap();

// Get weighted kernel matrix
let weighted = forgetful.get_weighted_matrix();

Implementations

impl ForgetfulKernelMatrix

pub fn new(kernel: Box<dyn Kernel>, config: ForgetfulConfig) -> Self

Create a new forgetful kernel matrix.

pub fn add_sample(&mut self, sample: Vec<f64>) -> Result<()>

Add a sample with forgetting.

pub fn get_matrix(&self) -> &Vec<Vec<f64>>

Get the raw (unweighted) kernel matrix.

pub fn get_weighted_matrix(&self) -> Vec<Vec<f64>>

Get the weighted kernel matrix.

Each entry K[i,j] is multiplied by sqrt(w_i * w_j) to maintain PSD property.

pub fn get_weights(&self) -> &Vec<f64>

Get sample weights.

pub fn get_samples(&self) -> &Vec<Vec<f64>>

Get the current samples.

pub fn len(&self) -> usize

Get the number of samples.

pub fn is_empty(&self) -> bool

Check if empty.

pub fn stats(&self) -> &OnlineStats

Get statistics.

pub fn lambda(&self) -> f64

Get the forgetting factor.

pub fn clear(&mut self)

Clear all samples.

pub fn compute_weighted(&self, query: &[f64]) -> Result<Vec<f64>>

Compute weighted kernel values between query and all stored samples.

pub fn effective_size(&self) -> f64

Get effective sample size (sum of weights).