Struct FuncScatterTensor 

pub struct FuncScatterTensor {
    pub in_scatter: Array2<f32>,
    pub out_scatter: Array2<f32>,
}

Represents a 4-tensor which is expressible as kron(out_scatter, in_scatter), where out_scatter and in_scatter are both positive semi-definite. This is typically used to express the precision or covariance of a crate::normal_inverse_wishart::NormalInverseWishart model. See also crate::func_schmear::FuncSchmear for an example of usage.

Fields

in_scatter: Array2<f32>

out_scatter: Array2<f32>

Implementations

impl FuncScatterTensor

pub fn inverse(&self) -> FuncScatterTensor

Inverts both in_scatter and out_scatter of this FuncScatterTensor to yield a new one. May be used to e.g: convert between a function’s covariance 4-tensor and its precision 4-tensor.

pub fn flatten_and_multiply(&self, mat: ArrayView2<'_, f32>) -> Array2<f32>

Supposing that out_scatter and in_scatter are nxn and mxm, respectively, and given a Lx(n*m) matrix mat, compute mat * kron(out_scatter, in_scatter). This has better asymptotic efficiency than simply expanding out the definition above. Delivers a result of size Lx(n*m).

pub fn compress(&self, mat: ArrayView2<'_, f32>) -> Array2<f32>

Efficiently transforms this FuncScatterTensor, taken to represent a covariance tensor, through the linear transformation given by mat.

Supposing that out_scatter and in_scatter are nxn and mxm, respectively, and given a Lx(n*m) matrix mat, compute mat * kron(out_scatter, in_scatter) * mat.t(). This has better asymptotic efficiency than simply expanding out the definition above. Delivers a result of size LxL.

pub fn flatten(&self) -> Array2<f32>

Flattens this FuncScatterTensor to a 2d matrix using the kronecker product kron(out_scatter, in_scatter).

Trait Implementations

impl Clone for FuncScatterTensor

fn clone(&self) -> FuncScatterTensor

Returns a duplicate of the value.

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

Performs copy-assignment from source.