Struct linfa_reduction::diffusion_map::DiffusionMapParams

pub struct DiffusionMapParams {
    pub steps: usize,
    pub embedding_size: usize,
}

Diffusion map hyperparameters

The diffusion map algorithms has only two explicit hyperparameter. The first is the stepsize. As the algorithm calculates the closeness of points after a number of steps taken in the diffusion graph, a larger step size introduces a more global behaviour of the projection while a smaller one (especially one) just projects close obserations closely together. The second parameter is the embedding size and defines the target dimensionality.

Fields

steps: usize

embedding_size: usize

Implementations

impl DiffusionMapParams

pub fn steps(self, steps: usize) -> Self

Set the number of steps in the diffusion operator

The diffusion map algorithm expresses the transition probability with a kernel matrix and then takes multiple steps along the diffusion operator. This scales in practice the eigenvalues of the decomposition exponentially with the number of steps.

pub fn validate(&self) -> Result<()>

Validates the parameter

Trait Implementations

impl<'a, F: Float> Transformer<&'a KernelBase<ArrayBase<OwnedRepr<F>, Dim<[usize; 2]>>, CsMatBase<F, usize, Vec<usize, Global>, Vec<usize, Global>, Vec<F, Global>, usize>>, Result<DiffusionMap<F>, Error>> for DiffusionMapParams

fn transform(&self, kernel: &'a Kernel<F>) -> Result<DiffusionMap<F>>

Project a kernel matrix to its embedding

Parameter

• kernel: Kernel matrix

Returns

Embedding for each observation in the kernel matrix