Struct fetish_lib::feature_space_info::FeatureSpaceInfo

pub struct FeatureSpaceInfo {
    pub base_dimensions: usize,
    pub feature_dimensions: usize,
    pub feature_collections: Vec<Box<dyn FeatureCollection>>,
    pub sketcher: Option<LinearSketch>,
}

Information about the base space dimension, the preferred sketcher for the space, the feature dimensions, and the feature mapping. In other words, this struct carries all information defining spaces which are associated to a particular type.

As in the FETISH paper, the way this is conceptually organized is: Base space -(sketcher)-> Compressed space -(feature mapping)-> Feature space.

Fields

base_dimensions: usize

feature_dimensions: usize

feature_collections: Vec<Box<dyn FeatureCollection>>

sketcher: Option<LinearSketch>

Implementations

impl FeatureSpaceInfo

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

Gets the projection matrix from the base space to the compressed space.

pub fn get_sketched_dimensions(&self) -> usize

Gets the dimensionality of the compressed space.

pub fn sketch(&self, mean: ArrayView1<'_, f32>) -> Array1<f32>

Transforms a vector from the base space to the compressed space.

pub fn compress_schmear(&self, schmear: &Schmear) -> Schmear

Transforms a schmear from the base space to the compressed space.

pub fn get_feature_jacobian(&self, in_vec: ArrayView1<'_, f32>) -> Array2<f32>

Gets the Jacobian for the feature mapping evaluated at the given compressed vector.

pub fn get_features_from_base(&self, in_vec: ArrayView1<'_, f32>) -> Array1<f32>

Maps a vector from the base space through the sketcher and through the feature map to yield a vector in the feature space.

pub fn get_features(&self, in_vec: ArrayView1<'_, f32>) -> Array1<f32>

Gets the feature vector for the given compressed vector.

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

See to_features_mat. Gets features for a collection of compressed vectors.

pub fn featurize_schmear(&self, x: &Schmear) -> Schmear

Given a Schmear in the compressed space, uses an unscented transform to estimate what it would map to through the feature mapping.