Struct ICMKernel 

pub struct ICMKernel { /* private fields */ }

Intrinsic Coregionalization Model (ICM) kernel.

K((x, i), (y, j)) = B[i, j] * k(x, y)

where:

• B is the task covariance matrix (num_tasks x num_tasks)
• k is the base kernel on features

This model assumes all tasks share the same underlying kernel but with different task-specific scales captured in B.

Implementations

impl ICMKernel

pub fn new( base_kernel: Box<dyn Kernel>, task_covariance: Vec<Vec<f64>>, ) -> Result<Self>

Create a new ICM kernel.

Arguments

• base_kernel - Kernel for feature similarity
• task_covariance - Positive semi-definite task covariance matrix

pub fn independent( base_kernel: Box<dyn Kernel>, num_tasks: usize, ) -> Result<Self>

Create ICM with identity task covariance (independent tasks).

pub fn uniform( base_kernel: Box<dyn Kernel>, num_tasks: usize, correlation: f64, ) -> Result<Self>

Create ICM with uniform task correlation.

pub fn from_rank1( base_kernel: Box<dyn Kernel>, task_variances: Vec<f64>, ) -> Result<Self>

Create ICM from rank-1 decomposition B = v * v^T.

This is useful when you have task-specific variances.

pub fn compute_tasks(&self, x: &TaskInput, y: &TaskInput) -> Result<f64>

Compute ICM kernel value for task inputs.

pub fn num_tasks(&self) -> usize

Get number of tasks.

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

Get task covariance matrix.

pub fn compute_task_matrix(&self, inputs: &[TaskInput]) -> Result<Vec<Vec<f64>>>

Compute full kernel matrix for multiple task inputs.